diff --git a/.gitignore b/.gitignore index a04655b2..a018ae6e 100644 --- a/.gitignore +++ b/.gitignore @@ -8,3 +8,5 @@ jsconfig.json node_modules/ go.sum .hugo_build.lock + +istruzioni.md \ No newline at end of file diff --git a/content/publication/baieri-2024-arap/cite.bib b/content/publication/baieri-2024-arap/cite.bib new file mode 100644 index 00000000..6a401b75 --- /dev/null +++ b/content/publication/baieri-2024-arap/cite.bib @@ -0,0 +1,9 @@ +@misc{baieri-2024-arap, + title={Implicit-ARAP: Efficient Handle-Guided Deformation of High-Resolution Meshes and Neural Fields via Local Patch Meshing}, + author={Daniele Baieri and Filippo Maggioli and Zorah Lähner and Simone Melzi and Emanuele Rodol\`a}, + year={2024}, + eprint={2405.12895}, + archivePrefix={arXiv}, + primaryClass={cs.GR}, + url={https://arxiv.org/abs/2405.12895}, +} \ No newline at end of file diff --git a/content/publication/baieri-2024-arap/featured.png b/content/publication/baieri-2024-arap/featured.png new file mode 100644 index 00000000..20d3b920 Binary files /dev/null and b/content/publication/baieri-2024-arap/featured.png differ diff --git a/content/publication/baieri-2024-arap/index.md b/content/publication/baieri-2024-arap/index.md new file mode 100644 index 00000000..fb415d98 --- /dev/null +++ b/content/publication/baieri-2024-arap/index.md @@ -0,0 +1,49 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'Implicit-ARAP: Efficient Handle-Guided Deformation of High-Resolution Meshes and Neural Fields via Local Patch Meshing' +subtitle: '' +summary: '' +authors: +- baieri +- maggioli +- Zorah Laehner +- melzi +- rodola +tags: [] +categories: [] +date: '2024-05-21' +lastmod: 2023-10-02T:26:44 +featured: false +draft: false +publication_short: "Preprint" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: 'Center' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-10-02T:26:44' +publication_types: +- '3' +abstract: "In this work, we present the local patch mesh representation for neural signed distance fields. This technique allows to discretize local regions of the level sets of an input SDF by projecting and deforming flat patch meshes onto the level set surface, using exclusively the SDF information and its gradient. Our analysis reveals this method to be more accurate than the standard marching cubes algorithm for approximating the implicit surface. Then, we apply this representation in the setting of handle-guided deformation: we introduce two distinct pipelines, which make use of 3D neural fields to compute As-Rigid-As-Possible deformations of both high-resolution meshes and neural fields under a given set of constraints. We run a comprehensive evaluation of our method and various baselines for neural field and mesh deformation which show both pipelines achieve impressive efficiency and notable improvements in terms of quality of results and robustness. With our novel pipeline, we introduce a scalable approach to solve a well-established geometry processing problem on high-resolution meshes, and pave the way for extending other geometric tasks to the domain of implicit surfaces via local patch meshing." +publication: '*arXiv preprint*' +links: +- name: arXiv + url : https://arxiv.org/abs/2405.12895 +- name: PDF + url: https://arxiv.org/pdf/2405.12895 +- icon: github + icon_pack: fab + name: 'GitHub' + url: https://github.com/daniele-baieri/implicit-arap +--- \ No newline at end of file diff --git a/content/publication/bonzi-2023-voice/bonzi-2023-voice.pdf b/content/publication/bonzi-2023-voice/bonzi-2023-voice.pdf new file mode 100644 index 00000000..b6cb6ba0 Binary files /dev/null and b/content/publication/bonzi-2023-voice/bonzi-2023-voice.pdf differ diff --git a/content/publication/bonzi-2023-voice/cite.bib b/content/publication/bonzi-2023-voice/cite.bib new file mode 100644 index 00000000..d8bd17e7 --- /dev/null +++ b/content/publication/bonzi-2023-voice/cite.bib @@ -0,0 +1,7 @@ +@article{bonzi-2023-voice, +author={Bonzi, Francesco and Mancusi, Michele and Deo, Simone Del and Melucci, Pierfrancesco and Tavella, Maria Stella and Parisi, Loreto and Rodol\`a, Emanuele}, +booktitle={2023 IEEE 33rd International Workshop on Machine Learning for Signal Processing (MLSP)}, +title={Exploiting Music Source Separation For Singing Voice Detection}, +year={2023}, +doi={10.1109/MLSP55844.2023.10285863} +} diff --git a/content/publication/bonzi-2023-voice/featured.png b/content/publication/bonzi-2023-voice/featured.png new file mode 100644 index 00000000..fdd1d81a Binary files /dev/null and b/content/publication/bonzi-2023-voice/featured.png differ diff --git a/content/publication/bonzi-2023-voice/index.md b/content/publication/bonzi-2023-voice/index.md new file mode 100644 index 00000000..6854c948 --- /dev/null +++ b/content/publication/bonzi-2023-voice/index.md @@ -0,0 +1,45 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: Exploiting Music Source Separation For Singing Voice Detection +subtitle: '' +summary: '' +authors: +- Francesco Bonzi +- mancusi +- Simone Del Deo +- Pierfrancesco Melucci +- Maria Stella Tavella +- Loreto Parisi +- rodola +tags: +- 'source separation' +- 'audio' +categories: [] +date: '2023-09-01' +lastmod: 2023-12-16T10:57:52+01:00 +featured: false +draft: false +publication_short: "MLSP" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: '' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-02-05T09:57:52.156096Z' +abstract: "Singing voice detection (SVD) is an essential task in many music information retrieval (MIR) applications. Deep learning methods have shown promising results for SVD, but further performance improvements are desirable since it underlies many other tasks. This work proposes a novel SVD system combining a state-of-the-art music source separator (Demucs) with two downstream models: Long-term Recurrent Convolutional Network (LRCN) and a Transformer network. Our work highlights two main aspects: the impact of a music source separation model, such as Demucs, and its zero-shot capabilities for the SVD task; and the potential for deep learning to improve the system’s performance further. We evaluate our approach on three datasets (Jamendo Corpus, MedleyDB, and MIR-IK) and compare the performance of the two models to a baseline root mean square (RMS) algorithm and the current state-of-the-art for the Jamendo Corpus dataset." +publication: '*International Workshop on Machine Learning for Signal Processing 2023*' +links: +- name: URL + url: https://ieeexplore.ieee.org/document/10285863 +--- diff --git a/content/publication/camoscio-23/cite.bib b/content/publication/camoscio-23/cite.bib new file mode 100644 index 00000000..c4ebac47 --- /dev/null +++ b/content/publication/camoscio-23/cite.bib @@ -0,0 +1,6 @@ +@inproceedings{camoscio-23, + title = {Camoscio: an Italian Instruction-tuned LLaMA}, + author = {Andrea Santilli and Emanuele Rodol{\`a}}, + booktitle = {Proc. CLiC-it}, + year = {2023}, +} \ No newline at end of file diff --git a/content/publication/camoscio-23/featured.png b/content/publication/camoscio-23/featured.png new file mode 100644 index 00000000..a5678830 Binary files /dev/null and b/content/publication/camoscio-23/featured.png differ diff --git a/content/publication/camoscio-23/index.md b/content/publication/camoscio-23/index.md new file mode 100644 index 00000000..f96754f5 --- /dev/null +++ b/content/publication/camoscio-23/index.md @@ -0,0 +1,54 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: "Camoscio: an Italian Instruction-tuned LLaMA" +subtitle: '' +summary: '' +authors: +- santilli +- rodola + +tags: +- LLM + +categories: [] +date: '2023-12-18' +lastmod: 2022-09-30T11:32:00+02:00 +featured: false +draft: false +publication_short: "CLiC-it 2023" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +# image: +# caption: '' +# focal_point: 'Center' +# preview_only: false + +links: +- name: PDF + url: https://ceur-ws.org/Vol-3596/paper44.pdf +- icon: github + icon_pack: fab + name: 'GitHub' + url: https://github.com/teelinsan/camoscio +- icon: award + icon_pack: fas + name: 'Best Student Paper Award' + url: https://clic2023.ilc.cnr.it/awards/ + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2022-04-28T10:30:59.888843Z' +publication_types: +- '1' + +abstract: "In recent years Large Language Models (LLMs) have increased the state of the art on several natural language processing tasks. However, their accessibility is often limited to paid API services, posing challenges for researchers in conducting extensive investigations. On the other hand, while some open-source models have been proposed by the community, they are typically English-centric or multilingual without a specific adaptation for the Italian language. In an effort to democratize the available and open resources for the Italian language, in this paper we introduce Camoscio: a language model specifically tuned to follow users' prompts in Italian. Specifically, we finetuned the smallest variant of LLaMA (7b) with LoRA on a corpus of instruction prompts translated to Italian via ChatGPT. Results indicate that the model's zero-shot performance on various downstream tasks in Italian competes favorably with existing models specifically finetuned for those tasks. All the artifacts (code, dataset, model) are released to the community." + +publication: '*Italian Conference on Computational Linguistics (CLiC-it 2023)*' +--- diff --git a/content/publication/cannistraci-2023-charts/cite.bib b/content/publication/cannistraci-2023-charts/cite.bib new file mode 100644 index 00000000..7983b894 --- /dev/null +++ b/content/publication/cannistraci-2023-charts/cite.bib @@ -0,0 +1,6 @@ +@inproceedings{cannistraci-2023-charts, + title={From Charts to Atlas: Merging Latent Spaces into One}, + author={Donato Crisostomi and Irene Cannistraci and Luca Moschella and Pietro Barbiero and Marco Ciccone and Pietro Li\`o and Emanuele Rodol\`a}, + year={2023}, + booktitle={Proc. NeurReps}, +} \ No newline at end of file diff --git a/content/publication/cannistraci-2023-charts/featured.png b/content/publication/cannistraci-2023-charts/featured.png new file mode 100644 index 00000000..8db54578 Binary files /dev/null and b/content/publication/cannistraci-2023-charts/featured.png differ diff --git a/content/publication/cannistraci-2023-charts/index.md b/content/publication/cannistraci-2023-charts/index.md new file mode 100644 index 00000000..539c8f44 --- /dev/null +++ b/content/publication/cannistraci-2023-charts/index.md @@ -0,0 +1,46 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'From Charts to Atlas: Merging Latent Spaces into One' +subtitle: '' +summary: '' +authors: +- crisostomi +- cannistraci +- moschella +- Pietro Barbiero +- Marco Ciccone +- Pietro Lio +- rodola +tags: +- 'Model merging' +categories: [] +date: '2023-11-29' +lastmod: 2023-12-16T10:57:52+01:00 +featured: false +draft: false +publication_short: "NeuReps 2023" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: '' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-02-05T09:57:52.156096Z' +abstract: "Models trained on semantically related datasets and tasks exhibit comparable inter-sample relations within their latent spaces. We investigate in this study the aggregation of such latent spaces to create a unified space encompassing the combined information. To this end, we introduce Relative Latent Space Aggregation (RLSA), a two-step approach that first renders the spaces comparable using relative representations, and then aggregates them via a simple mean. We carefully divide a classification problem into a series of learning tasks under three different settings: sharing samples, classes, or neither. We then train a model on each task and aggregate the resulting latent spaces. We compare the aggregated space with that derived from an end-to-end model trained over all tasks and show that the two spaces are similar. We then observe that the aggregated space is better suited for classification, and empirically demonstrate that it is due to the unique imprints left by task-specific embedders within the representations. We finally test our framework in scenarios where no shared region exists and show that it can still be used to merge the spaces, albeit with diminished benefits over naive merging." +publication: '*NeurReps Workshop 2023*' +links: +- name: URL + url: https://openreview.net/forum?id=ZFu7CPtznY +- name: PDF + url: https://openreview.net/pdf?id=ZFu7CPtznY +--- diff --git a/content/publication/cannistraci-2023-infusing/index.md b/content/publication/cannistraci-2023-infusing/index.md index 62823442..9c25e2d1 100644 --- a/content/publication/cannistraci-2023-infusing/index.md +++ b/content/publication/cannistraci-2023-infusing/index.md @@ -12,11 +12,11 @@ authors: - rodola tags: [] categories: [] -date: '2023-10-02' +date: '2024-06-02' lastmod: 2023-10-02T:26:44 featured: false draft: false -publication_short: "" +publication_short: "ICLR 2024" # Featured image # To use, add an image named `featured.jpg/png` to your page's folder. @@ -36,8 +36,14 @@ publishDate: '2023-10-02T:26:44' publication_types: - '3' abstract: 'It has been observed that representations learned by distinct neural networks conceal structural similarities when the models are trained under similar inductive biases. From a geometric perspective, identifying the classes of transformations and the related invariances that connect these representations is fundamental to unlocking applications, such as merging, stitching, and reusing different neural modules. However, estimating task-specific transformations a priori can be challenging and expensive due to several factors (e.g., weights initialization, training hyperparameters, or data modality). To this end, we introduce a versatile method to directly incorporate a set of invariances into the representations, constructing a product space of invariant components on top of the latent representations without requiring prior knowledge about the optimal invariance to infuse. We validate our solution on classification and reconstruction tasks, observing consistent latent similarity and downstream performance improvements in a zero-shot stitching setting. The experimental analysis comprises three modalities (vision, text, and graphs), twelve pretrained foundational models, eight benchmarks, and several architectures trained from scratch.' -publication: '*arXiv preprint arXiv:2206.06182*' +publication: '*International Conference on Learning Representations (ICLR 2024)*' links: - name: URL - url : https://arxiv.org/abs/2310.01211 ---- \ No newline at end of file + url : https://openreview.net/forum?id=vngVydDWft +- name: PDF + url: https://openreview.net/pdf?id=vngVydDWft +- icon: award + icon_pack: fas + name: 'ICLR 2024 spotlight' + url: https://iclr.cc/virtual/2024/poster/17521 +--- diff --git a/content/publication/ciranni-2024-cocola/cite.bib b/content/publication/ciranni-2024-cocola/cite.bib new file mode 100644 index 00000000..6fdef45b --- /dev/null +++ b/content/publication/ciranni-2024-cocola/cite.bib @@ -0,0 +1,9 @@ +@misc{ciranni-2024-cocola, + title={COCOLA: Coherence-Oriented Contrastive Learning of Musical Audio Representations}, + author={Ruben Ciranni and Emilian Postolache and Giorgio Mariani and Michele Mancusi and Luca Cosmo and Emanuele Rodol\`a}, + year={2024}, + eprint={2404.16969}, + archivePrefix={arXiv}, + primaryClass={cs.SD}, + url={https://arxiv.org/abs/2404.16969}, +} \ No newline at end of file diff --git a/content/publication/ciranni-2024-cocola/featured.png b/content/publication/ciranni-2024-cocola/featured.png new file mode 100644 index 00000000..dad3822f Binary files /dev/null and b/content/publication/ciranni-2024-cocola/featured.png differ diff --git a/content/publication/ciranni-2024-cocola/index.md b/content/publication/ciranni-2024-cocola/index.md new file mode 100644 index 00000000..6844635d --- /dev/null +++ b/content/publication/ciranni-2024-cocola/index.md @@ -0,0 +1,50 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'COCOLA: Coherence-Oriented Contrastive Learning of Musical Audio Representations' +subtitle: '' +summary: '' +authors: +- Ruben Ciranni +- postolache +- mariani +- mancusi +- cosmo +- rodola +tags: [] +categories: [] +date: '2024-04-29' +lastmod: 2023-10-02T:26:44 +featured: false +draft: false +publication_short: "Preprint" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: 'Center' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-10-02T:26:44' +publication_types: +- '3' +abstract: "We present COCOLA (Coherence-Oriented Contrastive Learning for Audio), a contrastive learning method for musical audio representations that captures the harmonic and rhythmic coherence between samples. Our method operates at the level of stems (or their combinations) composing music tracks and allows the objective evaluation of compositional models for music in the task of accompaniment generation. We also introduce a new baseline for compositional music generation called CompoNet, based on ControlNet, generalizing the tasks of MSDM, and quantify it against the latter using COCOLA. We release all models trained on public datasets containing separate stems (MUSDB18-HQ, MoisesDB, Slakh2100, and CocoChorales)." +publication: '*arXiv preprint*' +links: +- name: arXiv + url : https://arxiv.org/abs/2404.16969 +- name: PDF + url: https://arxiv.org/pdf/2404.16969 +- icon: github + icon_pack: fab + name: 'GitHub' + url: https://github.com/gladia-research-group/cocola +--- \ No newline at end of file diff --git a/content/publication/cosmo-2024-kernel/cite.bib b/content/publication/cosmo-2024-kernel/cite.bib new file mode 100644 index 00000000..a686e245 --- /dev/null +++ b/content/publication/cosmo-2024-kernel/cite.bib @@ -0,0 +1,8 @@ +@article{cosmo-2024-kernel, + author = {Cosmo, Luca and Minello, Giorgia and Bicciato, Alessandro and Bronstein, Michael and Rodolà, Emanuele and Rossi, Luca and Torsello, Andrea}, + title = {Graph Kernel Neural Networks}, + journal = {IEEE Transactions on Neural Networks and Learning Systems}, + year = {2024}, + abstract = {The convolution operator at the core of many modern neural architectures can effectively be seen as performing a dot product between an input matrix and a filter. While this is readily applicable to data such as images, which can be represented as regular grids in the Euclidean space, extending the convolution operator to work on graphs proves more challenging, due to their irregular structure. In this article, we propose to use graph kernels, i.e., kernel functions that compute an inner product on graphs, to extend the standard convolution operator to the graph domain. This allows us to define an entirely structural model that does not require computing the embedding of the input graph. Our architecture allows to plug-in any type of graph kernels and has the added benefit of providing some interpretability in terms of the structural masks that are learned during the training process, similar to what happens for convolutional masks in traditional convolutional neural networks (CNNs). We perform an extensive ablation study to investigate the model hyperparameters’ impact and show that our model achieves competitive performance on standard graph classification and regression datasets.} +} + diff --git a/content/publication/cosmo-2024-kernel/featured.png b/content/publication/cosmo-2024-kernel/featured.png new file mode 100644 index 00000000..e94ee246 Binary files /dev/null and b/content/publication/cosmo-2024-kernel/featured.png differ diff --git a/content/publication/cosmo-2024-kernel/index.md b/content/publication/cosmo-2024-kernel/index.md new file mode 100644 index 00000000..ecb9b1ba --- /dev/null +++ b/content/publication/cosmo-2024-kernel/index.md @@ -0,0 +1,48 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'Graph Kernel Neural Networks' +subtitle: '' +summary: '' +authors: +- cosmo +- Giorgia Minello +- Alessandro Bicciato +- Michael Bronstein +- rodola +- Luca Rossi +- Andrea Torsello +tags: +- 'Graph learning' +categories: [] +date: '2024-05-01' +lastmod: 2023-02-05T10:57:53+01:00 +featured: false +draft: false +publication_short: "TNNLS" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: '' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-02-05T09:57:52.951304Z' +publication_types: +- '2' +abstract: "The convolution operator at the core of many modern neural architectures can effectively be seen as performing a dot product between an input matrix and a filter. While this is readily applicable to data such as images, which can be represented as regular grids in the Euclidean space, extending the convolution operator to work on graphs proves more challenging, due to their irregular structure. In this article, we propose to use graph kernels, i.e., kernel functions that compute an inner product on graphs, to extend the standard convolution operator to the graph domain. This allows us to define an entirely structural model that does not require computing the embedding of the input graph. Our architecture allows to plug-in any type of graph kernels and has the added benefit of providing some interpretability in terms of the structural masks that are learned during the training process, similar to what happens for convolutional masks in traditional convolutional neural networks (CNNs). We perform an extensive ablation study to investigate the model hyperparameters’ impact and show that our model achieves competitive performance on standard graph classification and regression datasets." +publication: '*IEEE Transactions on Neural Networks and Learning Systems*' +links: +- name: URL + url: https://ieeexplore.ieee.org/document/10542111 +- name: arXiv + url: https://arxiv.org/abs/2112.07436 +--- diff --git a/content/publication/donato-2024-cycle/cite.bib b/content/publication/donato-2024-cycle/cite.bib new file mode 100644 index 00000000..ebc2c4df --- /dev/null +++ b/content/publication/donato-2024-cycle/cite.bib @@ -0,0 +1,9 @@ +@misc{donato-2024-cycle, + title={$C^2M^3$: Cycle-Consistent Multi-Model Merging}, + author={Donato Crisostomi and Marco Fumero and Daniele Baieri and Florian Bernard and Emanuele Rodol\`a}, + year={2024}, + eprint={2405.17897}, + archivePrefix={arXiv}, + primaryClass={cs.LG}, + url={https://arxiv.org/abs/2405.17897}, +} \ No newline at end of file diff --git a/content/publication/donato-2024-cycle/featured.png b/content/publication/donato-2024-cycle/featured.png new file mode 100644 index 00000000..805772ca Binary files /dev/null and b/content/publication/donato-2024-cycle/featured.png differ diff --git a/content/publication/donato-2024-cycle/index.md b/content/publication/donato-2024-cycle/index.md new file mode 100644 index 00000000..28bda358 --- /dev/null +++ b/content/publication/donato-2024-cycle/index.md @@ -0,0 +1,49 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'Cycle-Consistent Multi-Model Merging' +subtitle: '' +summary: '' +authors: +- crisostomi +- fumero +- baieri +- Florian Bernard +- rodola +tags: [] +categories: [] +date: '2024-05-28' +lastmod: 2023-10-02T:26:44 +featured: false +draft: false +publication_short: "Preprint" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: 'Center' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-10-02T:26:44' +publication_types: +- '3' +abstract: "In this paper, we present a novel data-free method for merging neural networks in weight space. Differently from most existing works, our method optimizes for the permutations of network neurons globally across all layers. This allows us to enforce cycle consistency of the permutations when merging N >= 3 models, allowing circular compositions of permutations to be computed without accumulating error along the path. We qualitatively and quantitatively motivate the need for such a constraint, showing its benefits when merging sets of models in scenarios spanning varying architectures and datasets. We finally show that, when coupled with activation renormalization, our approach yields the best results in the task." +publication: '*arXiv preprint*' +links: +- name: arXiv + url : https://arxiv.org/abs/2405.17897 +- name: PDF + url: https://arxiv.org/pdf/2405.17897 +- icon: github + icon_pack: fab + name: 'GitHub' + url: https://github.com/crisostomi/cycle-consistent-model-merging +--- \ No newline at end of file diff --git a/content/publication/eccv-24-rematch/cite.bib b/content/publication/eccv-24-rematch/cite.bib new file mode 100644 index 00000000..4c470e10 --- /dev/null +++ b/content/publication/eccv-24-rematch/cite.bib @@ -0,0 +1,8 @@ +@inproceedings{eccv-24-rematch, + abstract = {We introduce ReMatching, a novel shape correspondence solution based on the functional maps framework. Our method, by exploiting a new and appropriate re-meshing paradigm, can target shape-matching tasks even on meshes counting millions of vertices, where the original functional maps does not apply or requires a massive computational cost. The core of our procedure is a time-efficient remeshing algorithm which constructs a low-resolution geometry while acting conservatively on the original topology and metric. These properties allow translating the functional maps optimization problem on the resulting low-resolution representation, thus enabling efficient computation of correspondences with functional map approaches. Finally, we propose an efficient technique for extending the estimated correspondence to the original meshes. We show that our method is more efficient and effective through quantitative and qualitative comparisons, outperforming state-of-the-art pipelines in quality and computational cost.}, + author = {Maggioli, Filippo and Baieri, Daniele and Rodolà, Emanuele and Melzi, Simone}, + booktitle = {Proc. ECCV}, + pdf = {https://arxiv.org/pdf/2305.09274}, + title = {ReMatching: Low-Resolution Representations for Scalable Shape Correspondence}, + year = {2024} +} \ No newline at end of file diff --git a/content/publication/eccv-24-rematch/featured.png b/content/publication/eccv-24-rematch/featured.png new file mode 100644 index 00000000..b0fdc159 Binary files /dev/null and b/content/publication/eccv-24-rematch/featured.png differ diff --git a/content/publication/eccv-24-rematch/index.md b/content/publication/eccv-24-rematch/index.md new file mode 100644 index 00000000..6c387326 --- /dev/null +++ b/content/publication/eccv-24-rematch/index.md @@ -0,0 +1,41 @@ +--- + +title: "ReMatching: Low-Resolution Representations for Scalable Shape Correspondence" +subtitle: '' +summary: '' +authors: +- maggioli +- baieri +- rodola +- melzi + +tags: + - shape matching + - spectral methods + - functional maps + +categories: [] +date: '2024-07-05' +lastmod: 2024-07-05T16:50:39+01:00 +featured: false +draft: false + +image: + caption: '' + focal_point: '' + preview_only: false + +projects: [] +publishDate: '2024-07-05T15:50:38.894577Z' +publication_types: +- '1' + +abstract: We introduce ReMatching, a novel shape correspondence solution based on the functional maps framework. Our method, by exploiting a new and appropriate re-meshing paradigm, can target shape-matching tasks even on meshes counting millions of vertices, where the original functional maps does not apply or requires a massive computational cost. The core of our procedure is a time-efficient remeshing algorithm which constructs a low-resolution geometry while acting conservatively on the original topology and metric. These properties allow translating the functional maps optimization problem on the resulting low-resolution representation, thus enabling efficient computation of correspondences with functional map approaches. Finally, we propose an efficient technique for extending the estimated correspondence to the original meshes. We show that our method is more efficient and effective through quantitative and qualitative comparisons, outperforming state-of-the-art pipelines in quality and computational cost. + +publication: 'Proc. ECCV' +publication_short: "ECCV 2024" + +links: +- name: PDF + url: https://arxiv.org/pdf/2305.09274 +--- diff --git a/content/publication/fauno-23/cite.bib b/content/publication/fauno-23/cite.bib new file mode 100644 index 00000000..103cf9cc --- /dev/null +++ b/content/publication/fauno-23/cite.bib @@ -0,0 +1,6 @@ +@article{fauno-23, + title={Fauno: The Italian Large Language Model that will leave you senza parole!}, + author={Bacciu, Andrea and Trappolini, Giovanni and Santilli, Andrea and Rodol{\`a}, Emanuele and Silvestri, Fabrizio}, + journal={arXiv preprint arXiv:2306.14457}, + year={2023} +} \ No newline at end of file diff --git a/content/publication/fauno-23/featured.png b/content/publication/fauno-23/featured.png new file mode 100644 index 00000000..86002798 Binary files /dev/null and b/content/publication/fauno-23/featured.png differ diff --git a/content/publication/fauno-23/index.md b/content/publication/fauno-23/index.md new file mode 100644 index 00000000..577d7c2f --- /dev/null +++ b/content/publication/fauno-23/index.md @@ -0,0 +1,52 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: "Fauno - The Italian Large Language Model that will leave you senza parole!" +subtitle: '' +summary: '' +authors: +- Andrea Bacciu +- trappolini +- santilli +- rodola +- Fabrizio Silvestri + +tags: +- LLM + +categories: [] +date: '2023-06-18' +lastmod: 2022-09-30T11:32:00+02:00 +featured: false +draft: false +publication_short: "IIR 2023" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +# image: +# caption: '' +# focal_point: 'Center' +# preview_only: false + +links: +- name: PDF + url: https://ceur-ws.org/Vol-3448/paper-24.pdf +- icon: github + icon_pack: fab + name: 'GitHub' + url: https://github.com/RSTLess-research/Fauno-Italian-LLM + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2022-04-28T10:30:59.888843Z' +publication_types: +- '1' +abstract: "This paper presents Fauno, the first and largest open-source Italian conversational Large Language Model (LLM). Our goal with Fauno is to democratize the study of LLMs in Italian, demonstrating that obtaining a fine-tuned conversational bot with a single GPU is possible. In addition, we release a collection of datasets for conversational AI in Italian. The datasets on which we fine-tuned Fauno include various topics such as general question answering, computer science, and medical questions." + +publication: '*Italian Information Retrieval Workshop 2023*' +--- diff --git a/content/publication/frascaroli-2024-casper/cite.bib b/content/publication/frascaroli-2024-casper/cite.bib new file mode 100644 index 00000000..69c8fb53 --- /dev/null +++ b/content/publication/frascaroli-2024-casper/cite.bib @@ -0,0 +1,12 @@ +@article{frascaroli-2024-casper, +title = {Latent spectral regularization for continual learning}, +journal = {Pattern Recognition Letters}, +volume = {184}, +pages = {119-125}, +year = {2024}, +issn = {0167-8655}, +doi = {https://doi.org/10.1016/j.patrec.2024.06.020}, +author = {Emanuele Frascaroli and Riccardo Benaglia and Matteo Boschini and Luca Moschella and Cosimo Fiorini and Emanuele Rodolà and Simone Calderara}, +keywords = {Continual learning, Deep learning, Regularization, Spectral geometry, Incremental learning}, +abstract = {While biological intelligence grows organically as new knowledge is gathered throughout life, Artificial Neural Networks forget catastrophically whenever they face a changing training data distribution. Rehearsal-based Continual Learning (CL) approaches have been established as a versatile and reliable solution to overcome this limitation; however, sudden input disruptions and memory constraints are known to alter the consistency of their predictions. We study this phenomenon by investigating the geometric characteristics of the learner’s latent space and find that replayed data points of different classes increasingly mix up, interfering with classification. Hence, we propose a geometric regularizer that enforces weak requirements on the Laplacian spectrum of the latent space, promoting a partitioning behavior. Our proposal, called Continual Spectral Regularizer for Incremental Learning (CaSpeR-IL), can be easily combined with any rehearsal-based CL approach and improves the performance of SOTA methods on standard benchmarks.} +} \ No newline at end of file diff --git a/content/publication/frascaroli-2024-casper/featured.png b/content/publication/frascaroli-2024-casper/featured.png new file mode 100644 index 00000000..0f568201 Binary files /dev/null and b/content/publication/frascaroli-2024-casper/featured.png differ diff --git a/content/publication/frascaroli-2024-casper/index.md b/content/publication/frascaroli-2024-casper/index.md new file mode 100644 index 00000000..3142d673 --- /dev/null +++ b/content/publication/frascaroli-2024-casper/index.md @@ -0,0 +1,49 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'Latent spectral regularization for continual learning' +subtitle: '' +summary: '' +authors: +- Emanuele Frascaroli +- Riccardo Benaglia +- Matteo Boschini +- moschella +- Cosimo Fiorini +- rodola +- Simone Calderara +tags: +- 'continual learning' +- 'spectral methods' +categories: [] +date: '2024-04-01' +lastmod: 2023-02-05T10:57:53+01:00 +featured: false +draft: false +publication_short: "PRL" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: '' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-02-05T09:57:52.951304Z' +publication_types: +- '2' +abstract: "While biological intelligence grows organically as new knowledge is gathered throughout life, Artificial Neural Networks forget catastrophically whenever they face a changing training data distribution. Rehearsal-based Continual Learning (CL) approaches have been established as a versatile and reliable solution to overcome this limitation; however, sudden input disruptions and memory constraints are known to alter the consistency of their predictions. We study this phenomenon by investigating the geometric characteristics of the learner’s latent space and find that replayed data points of different classes increasingly mix up, interfering with classification. Hence, we propose a geometric regularizer that enforces weak requirements on the Laplacian spectrum of the latent space, promoting a partitioning behavior. Our proposal, called Continual Spectral Regularizer for Incremental Learning (CaSpeR-IL), can be easily combined with any rehearsal-based CL approach and improves the performance of SOTA methods on standard benchmarks." +publication: '*Pattern Recognition Letters*' +links: +- name: URL + url: https://www.sciencedirect.com/science/article/pii/S0167865524001909 +- name: arXiv + url: https://arxiv.org/abs/2301.03345 +--- diff --git a/content/publication/fumero-2024-fmaps/cite.bib b/content/publication/fumero-2024-fmaps/cite.bib new file mode 100644 index 00000000..8ccc2b06 --- /dev/null +++ b/content/publication/fumero-2024-fmaps/cite.bib @@ -0,0 +1,9 @@ +@misc{fumero-2024-fmaps, + title={Latent Functional Maps}, + author={Marco Fumero and Marco Pegoraro and Valentino Maiorca and Francesco Locatello and Emanuele Rodol\`a}, + year={2024}, + eprint={2406.14183}, + archivePrefix={arXiv}, + primaryClass={cs.LG}, + url={https://arxiv.org/abs/2406.14183}, +} \ No newline at end of file diff --git a/content/publication/fumero-2024-fmaps/featured.png b/content/publication/fumero-2024-fmaps/featured.png new file mode 100644 index 00000000..7f423c32 Binary files /dev/null and b/content/publication/fumero-2024-fmaps/featured.png differ diff --git a/content/publication/fumero-2024-fmaps/index.md b/content/publication/fumero-2024-fmaps/index.md new file mode 100644 index 00000000..47c334fa --- /dev/null +++ b/content/publication/fumero-2024-fmaps/index.md @@ -0,0 +1,45 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'Latent Functional Maps' +subtitle: '' +summary: '' +authors: +- fumero +- pegoraro +- maiorca +- Francesco Locatello +- rodola +tags: [] +categories: [] +date: '2024-06-21' +lastmod: 2023-10-02T:26:44 +featured: false +draft: false +publication_short: "Preprint" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: 'Center' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-10-02T:26:44' +publication_types: +- '3' +abstract: "Neural models learn data representations that lie on low-dimensional manifolds, yet modeling the relation between these representational spaces is an ongoing challenge. By integrating spectral geometry principles into neural modeling, we show that this problem can be better addressed in the functional domain, mitigating complexity, while enhancing interpretability and performances on downstream tasks. To this end, we introduce a multi-purpose framework to the representation learning community, which allows to: (i) compare different spaces in an interpretable way and measure their intrinsic similarity; (ii) find correspondences between them, both in unsupervised and weakly supervised settings, and (iii) to effectively transfer representations between distinct spaces. We validate our framework on various applications, ranging from stitching to retrieval tasks, demonstrating that latent functional maps can serve as a swiss-army knife for representation alignment." +publication: '*arXiv preprint*' +links: +- name: arXiv + url : https://arxiv.org/abs/2406.14183 +- name: PDF + url: https://arxiv.org/pdf/2406.14183 +--- \ No newline at end of file diff --git a/content/publication/fumero-neurips-23/cite.bib b/content/publication/fumero-neurips-23/cite.bib new file mode 100644 index 00000000..99da3007 --- /dev/null +++ b/content/publication/fumero-neurips-23/cite.bib @@ -0,0 +1,8 @@ +@inproceedings{fumero-neurips-23, + title = {Leveraging sparse and shared feature activations for disentangled representation learning}, + author = {Marco Fumero and Florian Wenzel and Luca Zancato and Alessandro Achille and Emanuele Rodol{\`a} and Stefano Soatto and Bernhard Sch{\:o}lkopf and Francesco Locatello}, + booktitle = {Proc. NeurIPS}, + year = {2023}, + url = {https://openreview.net/forum?id=IHR83ufYPy}, + abstract = {Recovering the latent factors of variation of high dimensional data has so far focused on simple synthetic settings. Mostly building on unsupervised and weakly-supervised objectives, prior work missed out on the positive implications for representation learning on real world data. In this work, we propose to leverage knowledge extracted from a diversified set of supervised tasks to learn a common disentangled representation. Assuming each supervised task only depends on an unknown subset of the factors of variation, we disentangle the feature space of a supervised multi-task model, with features activating sparsely across different tasks and information being shared as appropriate. Importantly, we never directly observe the factors of variations but establish that access to multiple tasks is sufficient for identifiability under sufficiency and minimality assumptions. We validate our approach on six real world distribution shift benchmarks, and different data modalities (images, text), demonstrating how disentangled representations can be transferred to real settings.} +} \ No newline at end of file diff --git a/content/publication/fumero-neurips-23/featured.png b/content/publication/fumero-neurips-23/featured.png new file mode 100644 index 00000000..a1dd919f Binary files /dev/null and b/content/publication/fumero-neurips-23/featured.png differ diff --git a/content/publication/fumero-neurips-23/index.md b/content/publication/fumero-neurips-23/index.md new file mode 100644 index 00000000..9d6aed3e --- /dev/null +++ b/content/publication/fumero-neurips-23/index.md @@ -0,0 +1,60 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: Leveraging sparse and shared feature activations for disentangled representation learning +subtitle: '' +summary: '' +authors: +- fumero +- Florian Wenzel +- Luca Zancato +- Alessandro Achille +- rodola +- Stefano Soatto +- Bernhard Scholkopf +- Francesco Locatello + +tags: +- representation learning +- disentanglement + +categories: [] +date: '2023-09-30' +lastmod: 2022-09-30T11:32:00+02:00 +featured: false +draft: false +publication_short: "NeurIPS 2023" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +# image: +# caption: '' +# focal_point: 'Center' +# preview_only: false + +links: +- icon: link + icon_pack: fas + name: 'URL' + url: https://openreview.net/forum?id=IHR83ufYPy +- name: PDF + url: https://openreview.net/pdf?id=IHR83ufYPy +- icon: award + icon_pack: fas + name: 'NeurIPS 2023 spotlight' + url: https://nips.cc/virtual/2023/poster/72132 + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2022-04-28T10:30:59.888843Z' +publication_types: +- '1' +abstract: 'Recovering the latent factors of variation of high dimensional data has so far focused on simple synthetic settings. Mostly building on unsupervised and weakly-supervised objectives, prior work missed out on the positive implications for representation learning on real world data. In this work, we propose to leverage knowledge extracted from a diversified set of supervised tasks to learn a common disentangled representation. Assuming each supervised task only depends on an unknown subset of the factors of variation, we disentangle the feature space of a supervised multi-task model, with features activating sparsely across different tasks and information being shared as appropriate. Importantly, we never directly observe the factors of variations but establish that access to multiple tasks is sufficient for identifiability under sufficiency and minimality assumptions. We validate our approach on six real world distribution shift benchmarks, and different data modalities (images, text), demonstrating how disentangled representations can be transferred to real settings.' + +publication: '*Conference on Neural Information Processing Systems (NeurIPS 2023)*' +--- diff --git a/content/publication/maggioli-2023-plant/cite.bib b/content/publication/maggioli-2023-plant/cite.bib new file mode 100644 index 00000000..ba5cb145 --- /dev/null +++ b/content/publication/maggioli-2023-plant/cite.bib @@ -0,0 +1,17 @@ +@inproceedings{maggioli-2023-plant, +author = {Maggioli, Filippo and Klein, Jonathan and H\"{a}drich, Torsten and Rodol\`{a}, Emanuele and Pa\l{}ubicki, Wojtek and Pirk, S\"{o}ren and Michels, Dominik L.}, +title = {A Physically-inspired Approach to the Simulation of Plant Wilting}, +year = {2023}, +isbn = {9798400703157}, +publisher = {Association for Computing Machinery}, +address = {New York, NY, USA}, +url = {https://doi.org/10.1145/3610548.3618218}, +doi = {10.1145/3610548.3618218}, +abstract = {Plants are among the most complex objects to be modeled in computer graphics. While a large body of work is concerned with structural modeling and the dynamic reaction to external forces, our work focuses on the dynamic deformation caused by plant internal wilting processes. To this end, we motivate the simulation of water transport inside the plant which is a key driver of the wilting process. We then map the change of water content in individual plant parts to branch stiffness values and obtain the wilted plant shape through a position based dynamics simulation. We show, that our approach can recreate measured wilting processes and does so with a higher fidelity than approaches ignoring the internal water flow. Realistic plant wilting is not only important in a computer graphics context but can also aid the development of machine learning algorithms in agricultural applications through the generation of synthetic training data.}, +booktitle = {SIGGRAPH Asia 2023 Conference Papers}, +articleno = {66}, +numpages = {8}, +keywords = {Diffusion Models, Digital Plants, Plant Modeling, Plant Wilting, Position Based Dynamics (PBD).}, +location = {Sydney, NSW, Australia}, +series = {SA '23} +} \ No newline at end of file diff --git a/content/publication/maggioli-2023-plant/featured.jpg b/content/publication/maggioli-2023-plant/featured.jpg new file mode 100644 index 00000000..4894cf97 Binary files /dev/null and b/content/publication/maggioli-2023-plant/featured.jpg differ diff --git a/content/publication/maggioli-2023-plant/index.md b/content/publication/maggioli-2023-plant/index.md new file mode 100644 index 00000000..c9378229 --- /dev/null +++ b/content/publication/maggioli-2023-plant/index.md @@ -0,0 +1,52 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'A Physically-inspired Approach to the Simulation of Plant Wilting' +subtitle: '' +summary: '' +authors: +- maggioli +- Jonathan Klein +- Torsten Hadrich +- rodola +- Wojtek Palubicki +- Soren Pirk +- Dominik Michels +tags: +- shape modeling +- simulation +categories: [] +date: '2023-12-01' +lastmod: 2023-02-08T15:02:32+01:00 +featured: false +draft: false +publication_short: "TOG" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: '' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-02-08T14:02:31.810661Z' +publication_types: +- '2' +abstract: "Plants are among the most complex objects to be modeled in computer graphics. While a large body of work is concerned with structural modeling and the dynamic reaction to external forces, our work focuses on the dynamic deformation caused by plant internal wilting processes. To this end, we motivate the simulation of water transport inside the plant which is a key driver of the wilting process. We then map the change of water content in individual plant parts to branch stiffness values and obtain the wilted plant shape through a position based dynamics simulation. We show, that our approach can recreate measured wilting processes and does so with a higher fidelity than approaches ignoring the internal water flow. Realistic plant wilting is not only important in a computer graphics context but can also aid the development of machine learning algorithms in agricultural applications through the generation of synthetic training data." +publication: '*ACM Trans. Graph.*' +doi: 10.1145/3610548.3618218 +links: +- icon: link + icon_pack: fas + name: 'URL' + url: https://dl.acm.org/doi/10.1145/3610548.3618218 +- name: PDF + url: https://dl.acm.org/doi/pdf/10.1145/3610548.3618218 +--- diff --git a/content/publication/maiorca-2024-inverse/cite.bib b/content/publication/maiorca-2024-inverse/cite.bib new file mode 100644 index 00000000..96dba9aa --- /dev/null +++ b/content/publication/maiorca-2024-inverse/cite.bib @@ -0,0 +1,9 @@ +@misc{maiorca-2024-inverse, + title={Latent Space Translation via Inverse Relative Projection}, + author={Valentino Maiorca and Luca Moschella and Marco Fumero and Francesco Locatello and Emanuele Rodol\`a}, + year={2024}, + eprint={2406.15057}, + archivePrefix={arXiv}, + primaryClass={cs.LG}, + url={https://arxiv.org/abs/2406.15057}, +} \ No newline at end of file diff --git a/content/publication/maiorca-2024-inverse/featured.png b/content/publication/maiorca-2024-inverse/featured.png new file mode 100644 index 00000000..e5566f56 Binary files /dev/null and b/content/publication/maiorca-2024-inverse/featured.png differ diff --git a/content/publication/maiorca-2024-inverse/index.md b/content/publication/maiorca-2024-inverse/index.md new file mode 100644 index 00000000..d2231d19 --- /dev/null +++ b/content/publication/maiorca-2024-inverse/index.md @@ -0,0 +1,45 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'Latent Space Translation via Inverse Relative Projection' +subtitle: '' +summary: '' +authors: +- maiorca +- moschella +- fumero +- Francesco Locatello +- rodola +tags: [] +categories: [] +date: '2024-06-21' +lastmod: 2023-10-02T:26:44 +featured: false +draft: false +publication_short: "Preprint" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: 'Center' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-10-02T:26:44' +publication_types: +- '3' +abstract: "The emergence of similar representations between independently trained neural models has sparked significant interest in the representation learning community, leading to the development of various methods to obtain communication between latent spaces. Latent space communication can be achieved in two ways: i) by independently mapping the original spaces to a shared or relative one; ii) by directly estimating a transformation from a source latent space to a target one. In this work, we combine the two into a novel method to obtain latent space translation through the relative space. By formalizing the invertibility of angle-preserving relative representations and assuming the scale invariance of decoder modules in neural models, we can effectively use the relative space as an intermediary, independently projecting onto and from other semantically similar spaces. Extensive experiments over various architectures and datasets validate our scale invariance assumption and demonstrate the high accuracy of our method in latent space translation. We also apply our method to zero-shot stitching between arbitrary pre-trained text and image encoders and their classifiers, even across modalities. Our method has significant potential for facilitating the reuse of models in a practical manner via compositionality." +publication: '*arXiv preprint*' +links: +- name: arXiv + url : https://arxiv.org/abs/2406.15057 +- name: PDF + url: https://arxiv.org/pdf/2406.15057 +--- \ No newline at end of file diff --git a/content/publication/marin-smooth-22/cite.bib b/content/publication/marin-smooth-22/cite.bib new file mode 100644 index 00000000..962acd04 --- /dev/null +++ b/content/publication/marin-smooth-22/cite.bib @@ -0,0 +1,7 @@ +@article{marin-smooth-22, + TITLE = {Smoothness and effective regularizations in learned embeddings for shape matching}, + AUTHOR = {Marin, Riccardo and Attaiki, Souhaib and Melzi, Simone and Rodol{\`a}, Emanuele and Ovsjanikov, Maks}, + YEAR = {2022}, + journal = {CoRR}, + volume = {abs/2112.07289}, +} \ No newline at end of file diff --git a/content/publication/marin-smooth-22/featured.png b/content/publication/marin-smooth-22/featured.png new file mode 100644 index 00000000..7e8a8a29 Binary files /dev/null and b/content/publication/marin-smooth-22/featured.png differ diff --git a/content/publication/marin-smooth-22/index.md b/content/publication/marin-smooth-22/index.md new file mode 100644 index 00000000..d1197086 --- /dev/null +++ b/content/publication/marin-smooth-22/index.md @@ -0,0 +1,45 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'Smoothness and effective regularizations in learned embeddings for shape matching' +subtitle: '' +summary: '' +authors: +- marin +- Souhaib Attaiki +- melzi +- rodola +- Maks Ovsjanikov +tags: [] +categories: [] +date: '2022-06-02' +lastmod: 2023-10-02T:26:44 +featured: false +draft: false +publication_short: "Preprint" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: 'Center' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-10-02T:26:44' +publication_types: +- '3' +abstract: 'Many innovative applications require establishing correspondences among 3D geometric objects. However, the countless possible deformations of smooth surfaces make shape matching a challenging task. Finding an embedding to represent the different shapes in high-dimensional space where the matching is easier to solve is a well-trodden path that has given many outstanding solutions. Recently, a new trend has shown advantages in learning such representations. This novel idea motivated us to investigate which properties differentiate these data-driven embeddings and which ones promote state-of-the-art results. In this study, we analyze, for the first time, properties that arise in data-driven learned embedding and their relation to the shape-matching task. Our discoveries highlight the close link between matching and smoothness, which naturally emerge from training. Also, we demonstrate the relation between the orthogonality of the embedding and the bijectivity of the correspondence. Our experiments show exciting results, overcoming well-established alternatives and shedding a different light on relevant contexts and properties for learned embeddings.' +publication: '*arXiv preprint*' +links: +- name: URL + url : https://arxiv.org/abs/2112.07289 +- name: PDF + url: https://arxiv.org/pdf/2112.07289 +--- \ No newline at end of file diff --git a/content/publication/melzi-19/index.md b/content/publication/melzi-19/index.md index 1a7c164a..0fa48d5c 100644 --- a/content/publication/melzi-19/index.md +++ b/content/publication/melzi-19/index.md @@ -20,7 +20,7 @@ date: '2019-11-01' lastmod: 2023-02-08T15:02:32+01:00 featured: false draft: false -publication_short: "" +publication_short: "TOG" # Featured image # To use, add an image named `featured.jpg/png` to your page's folder. @@ -29,12 +29,6 @@ image: caption: '' focal_point: '' preview_only: false - -links: - - name: URL - url: http://www.lix.polytechnique.fr/~maks/papers/SGA19_zoomOut_reduced.pdf - - name: GitHub - url: https://github.com/llorz/SGA19_zoomOut # Projects (optional). # Associate this post with one or more of your projects. @@ -45,22 +39,18 @@ projects: [] publishDate: '2023-02-08T14:02:31.810661Z' publication_types: - '2' -abstract: We present a simple and efficient method for refining maps or correspondences - by iterative upsampling in the spectral domain that can be implemented in a few - lines of code. Our main observation is that high quality maps can be obtained even - if the input correspondences are noisy or are encoded by a small number of coefficients - in a spectral basis. We show how this approach can be used in conjunction with existing - initialization techniques across a range of application scenarios, including symmetry - detection, map refinement across complete shapes, non-rigid partial shape matching - and function transfer. In each application we demonstrate an improvement with respect - to both the quality of the results and the computational speed compared to the best - competing methods, with up to two orders of magnitude speed-up in some applications. - We also demonstrate that our method is both robust to noisy input and is scalable - with respect to shape complexity. Finally, we present a theoretical justification - for our approach, shedding light on structural properties of functional maps. +abstract: "We present a simple and efficient method for refining maps or correspondences by iterative upsampling in the spectral domain that can be implemented in a few lines of code. Our main observation is that high quality maps can be obtained even if the input correspondences are noisy or are encoded by a small number of coefficients in a spectral basis. We show how this approach can be used in conjunction with existing initialization techniques across a range of application scenarios, including symmetry detection, map refinement across complete shapes, non-rigid partial shape matching and function transfer. In each application we demonstrate an improvement with respect to both the quality of the results and the computational speed compared to the best competing methods, with up to two orders of magnitude speed-up in some applications. We also demonstrate that our method is both robust to noisy input and is scalable with respect to shape complexity. Finally, we present a theoretical justification for our approach, shedding light on structural properties of functional maps." publication: '*ACM Trans. Graph.*' doi: 10.1145/3355089.3356524 links: -- name: URL - url: https://doi.org/10.1145/3355089.3356524 +- icon: link + icon_pack: fas + name: 'URL' + url: https://dl.acm.org/doi/10.1145/3355089.3356524 +- name: PDF + url: https://arxiv.org/pdf/1904.07865 +- icon: github + icon_pack: fab + name: 'GitHub' + url: https://github.com/llorz/SGA19_zoomOut --- diff --git a/content/publication/palandra-2024-gsedit/cite.bib b/content/publication/palandra-2024-gsedit/cite.bib new file mode 100644 index 00000000..5ebf89cf --- /dev/null +++ b/content/publication/palandra-2024-gsedit/cite.bib @@ -0,0 +1,9 @@ +@misc{palandra-2024-gsedit, + title={GSEdit: Efficient Text-Guided Editing of 3D Objects via Gaussian Splatting}, + author={Francesco Palandra and Andrea Sanchietti and Daniele Baieri and Emanuele Rodol\`a}, + year={2024}, + eprint={2403.05154}, + archivePrefix={arXiv}, + primaryClass={cs.CV}, + url={https://arxiv.org/abs/2403.05154}, +} \ No newline at end of file diff --git a/content/publication/palandra-2024-gsedit/featured.png b/content/publication/palandra-2024-gsedit/featured.png new file mode 100644 index 00000000..6b1e04dd Binary files /dev/null and b/content/publication/palandra-2024-gsedit/featured.png differ diff --git a/content/publication/palandra-2024-gsedit/index.md b/content/publication/palandra-2024-gsedit/index.md new file mode 100644 index 00000000..00f10fc5 --- /dev/null +++ b/content/publication/palandra-2024-gsedit/index.md @@ -0,0 +1,44 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'GSEdit: Efficient Text-Guided Editing of 3D Objects via Gaussian Splatting' +subtitle: '' +summary: '' +authors: +- Francesco Palandra +- Andrea Sanchietti +- baieri +- rodola +tags: [] +categories: [] +date: '2024-05-21' +lastmod: 2023-10-02T:26:44 +featured: false +draft: false +publication_short: "Preprint" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: 'Center' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-10-02T:26:44' +publication_types: +- '3' +abstract: "We present GSEdit, a pipeline for text-guided 3D object editing based on Gaussian Splatting models. Our method enables the editing of the style and appearance of 3D objects without altering their main details, all in a matter of minutes on consumer hardware. We tackle the problem by leveraging Gaussian splatting to represent 3D scenes, and we optimize the model while progressively varying the image supervision by means of a pretrained image-based diffusion model. The input object may be given as a 3D triangular mesh, or directly provided as Gaussians from a generative model such as DreamGaussian. GSEdit ensures consistency across different viewpoints, maintaining the integrity of the original object's information. Compared to previously proposed methods relying on NeRF-like MLP models, GSEdit stands out for its efficiency, making 3D editing tasks much faster. Our editing process is refined via the application of the SDS loss, ensuring that our edits are both precise and accurate. Our comprehensive evaluation demonstrates that GSEdit effectively alters object shape and appearance following the given textual instructions while preserving their coherence and detail." +publication: '*arXiv preprint*' +links: +- name: URL + url : https://arxiv.org/abs/2403.05154 +- name: PDF + url: https://arxiv.org/pdf/2403.05154 +--- \ No newline at end of file diff --git a/content/publication/pegoraro-2023-spectral/index.md b/content/publication/pegoraro-2023-spectral/index.md index 4132625f..9cba974c 100644 --- a/content/publication/pegoraro-2023-spectral/index.md +++ b/content/publication/pegoraro-2023-spectral/index.md @@ -21,7 +21,7 @@ date: '2023-01-01' lastmod: 2023-12-16T10:57:52+01:00 featured: false draft: false -publication_short: "NeuReps" +publication_short: "NeuReps 2023" # Featured image # To use, add an image named `featured.jpg/png` to your page's folder. @@ -55,8 +55,14 @@ abstract: In graph learning, maps between graphs and their subgraphs frequently Our approach bears practical benefits in knowledge distillation and hierarchical learning, where we show comparable or improved performance at a fraction of the computational cost. -publication: '*NeuReps Workshop 2023*' +publication: '*NeurReps Workshop 2023*' links: - name: URL - url: https://arxiv.org/pdf/2205.14938.pdf + url: https://openreview.net/forum?id=e9JBa515z2 +- name: PDF + url: https://openreview.net/pdf?id=e9JBa515z2 +- icon: award + icon_pack: fas + name: 'Best Paper Award' + url: https://www.neurreps.org/ --- diff --git a/content/publication/postolache-2024-eeg/cite.bib b/content/publication/postolache-2024-eeg/cite.bib new file mode 100644 index 00000000..ff52a25d --- /dev/null +++ b/content/publication/postolache-2024-eeg/cite.bib @@ -0,0 +1,8 @@ +@misc{postolache-2024-eeg, + title={Naturalistic Music Decoding from EEG Data via Latent Diffusion Models}, + author={Emilian Postolache and Natalia Polouliakh and Hiroaki Kitano and Akima Connelly and Emanuele Rodol\`a and Luca Cosmo and Taketo Akama}, + year={2024}, + eprint={2405.09062}, + archivePrefix={arXiv}, + url={https://arxiv.org/abs/2405.09062} +} \ No newline at end of file diff --git a/content/publication/postolache-2024-eeg/featured.png b/content/publication/postolache-2024-eeg/featured.png new file mode 100644 index 00000000..da7c0d8d Binary files /dev/null and b/content/publication/postolache-2024-eeg/featured.png differ diff --git a/content/publication/postolache-2024-eeg/index.md b/content/publication/postolache-2024-eeg/index.md new file mode 100644 index 00000000..8fcb5fd6 --- /dev/null +++ b/content/publication/postolache-2024-eeg/index.md @@ -0,0 +1,47 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'Naturalistic Music Decoding from EEG Data via Latent Diffusion Models' +subtitle: '' +summary: '' +authors: +- postolache +- Natalia Polouliakh +- Hiroaki Kitano +- Akima Connelly +- rodola +- cosmo +- Taketo Akama +tags: [] +categories: [] +date: '2024-07-02' +lastmod: 2023-10-02T:26:44 +featured: false +draft: false +publication_short: "Preprint" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: 'Center' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-10-02T:26:44' +publication_types: +- '3' +abstract: "In this article, we explore the potential of using latent diffusion models, a family of powerful generative models, for the task of reconstructing naturalistic music from electroencephalogram (EEG) recordings. Unlike simpler music with limited timbres, such as MIDI-generated tunes or monophonic pieces, the focus here is on intricate music featuring a diverse array of instruments, voices, and effects, rich in harmonics and timbre. This study represents an initial foray into achieving general music reconstruction of high-quality using non-invasive EEG data, employing an end-to-end training approach directly on raw data without the need for manual pre-processing and channel selection. We train our models on the public NMED-T dataset and perform quantitative evaluation proposing neural embedding-based metrics. We additionally perform song classification based on the generated tracks. Our work contributes to the ongoing research in neural decoding and brain-computer interfaces, offering insights into the feasibility of using EEG data for complex auditory information reconstruction." +publication: '*arXiv preprint*' +links: +- name: URL + url : https://arxiv.org/abs/2405.09062 +- name: PDF + url: https://arxiv.org/pdf/2405.09062 +--- \ No newline at end of file diff --git a/content/publication/ricciardi-2024-stitch/cite.bib b/content/publication/ricciardi-2024-stitch/cite.bib new file mode 100644 index 00000000..c9a8606e --- /dev/null +++ b/content/publication/ricciardi-2024-stitch/cite.bib @@ -0,0 +1,6 @@ +@article{ricciardi-2024-stitch, + title={Zero-Shot Stitching in Reinforcement Learning using Relative Representations}, + author={Antonio Pio Ricciardi and Valentino Maiorca and Luca Moschella and Riccardo Marin and Emanuele Rodol\`a}, + year={2024}, + booktitle={European Workshop on Reinforcement Learning} +} \ No newline at end of file diff --git a/content/publication/ricciardi-2024-stitch/featured.png b/content/publication/ricciardi-2024-stitch/featured.png new file mode 100644 index 00000000..f289d720 Binary files /dev/null and b/content/publication/ricciardi-2024-stitch/featured.png differ diff --git a/content/publication/ricciardi-2024-stitch/index.md b/content/publication/ricciardi-2024-stitch/index.md new file mode 100644 index 00000000..28a484e0 --- /dev/null +++ b/content/publication/ricciardi-2024-stitch/index.md @@ -0,0 +1,45 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'Zero-shot stitching in Reinforcement Learning using Relative Representations' +subtitle: '' +summary: '' +authors: +- ricciardi +- maiorca +- moschella +- marin +- rodola +tags: +- 'Reinforcement learning' +- 'Zero-shot' +categories: [] +date: '2023-09-01' +lastmod: 2023-12-16T10:57:52+01:00 +featured: false +draft: false +publication_short: "EWRL 2023" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: '' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-02-05T09:57:52.156096Z' +abstract: "Visual Reinforcement Learning is a popular and powerful framework that takes full advantage of the Deep Learning breakthrough. However, it is also known that variations in the input (e.g., different colors of the panorama due to the season of the year) or the task (e.g., changing the speed limit for a car to respect) could require complete retraining of the agents. In this work, we leverage recent developments in unifying latent representations to demonstrate that it is possible to combine the components of an agent, rather than retrain it from scratch. We build upon the recent relative representations framework and adapt it for Visual RL. This allows us to create completely new agents capable of handling environment-task combinations never seen during training. Our work paves the road toward a more accessible and flexible use of reinforcement learning." +publication: '*European Workshop on Reinforcement Learning 2024*' +links: +- name: URL + url: https://openreview.net/forum?id=4tcXsImfsS1 +- name: PDF + url: https://openreview.net/pdf?id=4tcXsImfsS1 +--- diff --git a/content/publication/yu-2023-duet/cite.bib b/content/publication/yu-2023-duet/cite.bib new file mode 100644 index 00000000..5c5ce4b9 --- /dev/null +++ b/content/publication/yu-2023-duet/cite.bib @@ -0,0 +1,6 @@ +@inproceedings{yu-2023-duet, + title={Zero-Shot Duet Singing Voices Separation with Diffusion Models}, + author={Chin-Yun Yu and Emilian Postolache and Emanuele Rodol\`a and Gyorgy Fazekas}, + year={2023}, + booktitle={Sound Demixing Workshop (SDX)} +} \ No newline at end of file diff --git a/content/publication/yu-2023-duet/featured.png b/content/publication/yu-2023-duet/featured.png new file mode 100644 index 00000000..89189353 Binary files /dev/null and b/content/publication/yu-2023-duet/featured.png differ diff --git a/content/publication/yu-2023-duet/index.md b/content/publication/yu-2023-duet/index.md new file mode 100644 index 00000000..ca3c0fad --- /dev/null +++ b/content/publication/yu-2023-duet/index.md @@ -0,0 +1,48 @@ +--- +# Documentation: https://wowchemy.com/docs/managing-content/ + +title: 'Zero-Shot Duet Singing Voices Separation with Diffusion Models' +subtitle: '' +summary: '' +authors: +- Chin-Yun Yu +- postolache +- rodola +- Gyorgy Fazekas +tags: [] +categories: [] +date: '2023-11-04' +lastmod: 2023-10-02T:26:44 +featured: false +draft: false +publication_short: "SDX 2023" + +# Featured image +# To use, add an image named `featured.jpg/png` to your page's folder. +# Focal points: Smart, Center, TopLeft, Top, TopRight, Left, Right, BottomLeft, Bottom, BottomRight. +image: + caption: '' + focal_point: 'Center' + preview_only: false + +# Projects (optional). +# Associate this post with one or more of your projects. +# Simply enter your project's folder or file name without extension. +# E.g. `projects = ["internal-project"]` references `content/project/deep-learning/index.md`. +# Otherwise, set `projects = []`. +projects: [] +publishDate: '2023-10-02T:26:44' +publication_types: +- '3' +abstract: "In recent studies, diffusion models have shown promise as priors for solving audio inverse problems, including source separation. These models allow us to sample from the posterior distribution of a target signal given an observed signal by manipulating the diffusion process. However, when separating audio sources of the same type, such as duet singing voices, the prior learned by the diffusion process may not be sufficient to maintain the consistency of the source identity in the separated audio. For example, the singer may change from one to another from time to time. Tackling this problem will be useful for separating sources in a choir, or a mixture of multiple instruments with similar timbre, without acquiring large amounts of paired data. In this paper, we examine this problem in the context of duet singing voices separation, and propose a method to enforce the coherency of singer identity by splitting the mixture into overlapping segments and performing posterior sampling in an auto-regressive manner, conditioning on the previous segment. We evaluate the proposed method on the MedleyVox dataset with different overlap ratios, and show that the proposed method outperforms naive posterior sampling baseline. Our source code and the pre-trained model are publicly available." +publication: '*Sound Demixing Workshop 2023*' +links: +- name: PDF + url: https://sdx-workshop.github.io/papers/Yu.pdf +- name: arXiv + url: https://arxiv.org/abs/2311.07345 +- icon: github + icon_pack: fab + name: 'GitHub' + url: https://github.com/yoyololicon/duet-svs-diffusion +--- \ No newline at end of file