Dynamics-Aware Theory of Deep Learning

Información del proyecto

DYNASTY

Identificador del acuerdo de subvención: 101039676

DOI

10.3030/101039676

Fecha de la firma de la CE 3 Junio 2022

Fecha de inicio 1 Octubre 2022

Fecha de finalización 30 Septiembre 2027

Financiado con arreglo a

• European Research Council (ERC)

Coste total

€ 1 498 410,00

Aportación de la UE

€ 1 498 410,00

Coordinado por

INSTITUT NATIONAL DE RECHERCHE EN INFORMATIQUE ET AUTOMATIQUE
France

CORDIS proporciona enlaces a los documentos públicos y las publicaciones de los proyectos de los programas marco HORIZONTE.

Los enlaces a los documentos y las publicaciones de los proyectos del Séptimo Programa Marco, así como los enlaces a algunos tipos de resultados específicos, como conjuntos de datos y «software», se obtienen dinámicamente de OpenAIRE .

Resultado final

Documents, reports (1)

Data Management Plan

Drafting of Data Management Plan

Publicaciones

Conference proceedings (15)

Implicit Compressibility of Overparametrized Neural Networks Trained with Heavy-Tailed SGD

Autores: Wan, Yijun; Barsbey, Melih; Şimşekli, Umut; Zaidi, Abdellatif
Publicado en: The Thirty-Seventh Annual Conference on Neural Information Processing Systems (NeurIPS), Dec 2023, New Orleans, United States, 2023
Editor: PMLR
DOI: 10.48550/arxiv.2306.08125

Generalization Bounds for Heavy-Tailed SDEs through the Fractional Fokker-Planck Equation

Autores: Benjamin Dupuis, Umut Simsekli
Publicado en: NeurIPS, 2023
Editor: PMLR
DOI: 10.48550/arxiv.2402.07723

Approximate Heavy Tails in Offline (Multi-Pass) Stochastic Gradient Descent

Autores: Krunoslav Pavasovic, Alain Durmus, Umut Simsekli
Publicado en: NeurIPS, 2023
Editor: PMLR
DOI: 10.48550/arxiv.2310.18455

From Mutual Information to Expected Dynamics: New Generalization Bounds for Heavy-Tailed SGD

Autores: Benjamin Dupuis, Paul Viallard
Publicado en: NeurIPS Workshop on Heavy Tails in Machine Learning, 2023
Editor: Openreview

Chaotic Regularization and Heavy-Tailed Limits for Deterministic Gradient Descent

Autores: Lim, Soon Hoe; Wan, Yijun; Şimşekli, Umut
Publicado en: NeurIPS 2022 - Thirty-sixth Conference on Neural Information Processing Systems, 2022
Editor: PMLR
DOI: 10.48550/arxiv.2205.11361

Piecewise deterministic generative models

Autores: Andrea Bertazzi, Dario Shariatian, Umut Simsekli, Eric Moulines, Alain Durmus
Publicado en: NeurIPS, 2024
Editor: PMLR
DOI: 10.48550/arxiv.2407.19448

Algorithmic Stability of Heavy-Tailed SGD with General Loss Functions

Autores: Raj, Anant; Zhu, Lingjiong; Gürbüzbalaban, Mert; Şimşekli, Umut
Publicado en: ICML, 2023
Editor: PMLR
DOI: 10.48550/arxiv.2301.11885

Learning via Wasserstein-Based High Probability Generalisation Bounds

Autores: Viallard, Paul; Haddouche, Maxime; Şimşekli, Umut; Guedj, Benjamin
Publicado en: NeurIPS 2023, 2023
Editor: PMLR
DOI: 10.48550/arxiv.2306.04375

Generalization Bounds with Data-dependent Fractal Dimensions

Autores: Dupuis, Benjamin; Deligiannidis, George; Şimşekli, Umut
Publicado en: ICML, 2023
Editor: PMLR
DOI: 10.48550/arxiv.2302.02766

Generalization Guarantees via Algorithm-dependent Rademacher Complexity

Autores: Sachs, Sarah; van Erven, Tim; Hodgkinson, Liam; Khanna, Rajiv; Şimşekli, Umut
Publicado en: COLT 2023 - 36th Annual Conference on Learning Theory, 2023, Bangalore (Virtual event), India, 2023
Editor: PMLR
DOI: 10.48550/arxiv.2307.02501

Topological Generalization Bounds for Discrete-Time Stochastic Optimization Algorithms

Autores: Rayna Andreeva, Benjamin Dupuis, Rik Sarkar, Tolga Birdal, Umut Simsekli
Publicado en: NeurIPS, 2024
Editor: PMLR
DOI: 10.48550/arxiv.2407.08723

Uniform-in-Time Wasserstein Stability Bounds for (Noisy) Stochastic Gradient Descent

Autores: Zhu, Lingjiong; Gurbuzbalaban, Mert; Raj, Anant; Şimşekli, Umut
Publicado en: NeurIPS 2023 - Thirty-seventh Annual Conference on Neural Information Processing Systems,, 2023
Editor: PMLR
DOI: 10.48550/arxiv.2305.12056

Generalization Bounds for Stochastic Gradient Descent via Localized $\varepsilon$-Covers

Autores: Park, Sejun; Şimşekli, Umut; Erdogdu, Murat A.
Publicado en: NeurIPS 2022 - Thirty-sixth Conference on Neural Information Processing Systems, 2022, ISSN 1049-5258
Editor: PMLR
DOI: 10.48550/arxiv.2209.08951

Efficient Sampling of Stochastic Differential Equations with Positive Semi-Definite Models

Autores: Raj, Anant; Şimşekli, Umut; Rudi, Alessandro
Publicado en: NeurIPS 2023 - Thirty-seventh Annual Conference on Neural Information Processing Systems, Dec 2023, New Orleans, United States, 2023
Editor: PMLR
DOI: 10.48550/arxiv.2303.17109

Algorithmic Stability of Heavy-Tailed Stochastic Gradient Descent on Least Squares

Autores: Raj, Anant; Barsbey, Melih; Gürbüzbalaban, Mert; Zhu, Lingjiong; Şimşekli, Umut
Publicado en: ALT 2023 - 34th International Conference on Algorithmic Learning Theory, Feb 2023, Singapore, Singapore, 2022
Editor: PMLR
DOI: 10.48550/arxiv.2206.01274

Peer reviewed articles (1)

Cyclic and Randomized Stepsizes Invoke Heavier Tails in SGD than Constant Stepsize

Autores: Gürbüzbalaban, Mert; Hu, Yuanhan; Şimşekli, Umut; Zhu, Lingjiong
Publicado en: TMLR, 2023, ISSN 2835-8856
Editor: PMLR
DOI: 10.48550/arxiv.2302.05516