El objetivo del seminario es discutir algunos puntos de vista vigentes sobre la relación entre modelos de optimización regularizada (tipo LASSO) y la teoría matemática de aprendizaje automático (machine learning). El seminario será autocontenido y no asume conocimientos previos ni de optimización ni de ML.
HORARIO: Jueves 14.30-16.00 en Salón 703-Rojo FIng.
CALENDARIO DEL SEMINARIO:
Fecha | Titulo | Tema | Conferencista | Links | |
29/8 | Métrica de Wasserstein y machine learning distribucionalmente robusto (DRO) | DRO | M.Velasco | [arXiv], [Video] | |
5/9 | Suspendido | ||||
12/9 | Construcción de Matrices para Compressive Sensing | P.Raigorodsky | [Donoho, Bruckstein, Elad] | ||
19/9 | Landscape de optimización cuando factorizamos matrices | M.Fiori | [arXiv] | ||
26/9 | Minimización de rango mediante la norma nuclear | F.Carrasco | [arXiv] | ||
3/10 | Universal priors for sparse coding | I.Ramírez | [arXiv] | ||
10/10 | Brief history of denoising and their “nobel” conception as implicit manifold learners. | M. Di Martino | [slides][arXiv], [arXiv] | ||
17/10 | No hay seminario | ||||
24/10 | On the Benefits of Rank in Attention Layers | L. Raad | [arXiv] | ||
31/10 | No hay seminario | ||||
7/11 | Conformal Prediction | B. Marenco | [slides] [notas] | ||
14/11 | El Kernel de CD para Análisis de Datos | L. Bentancur | [slides] | ||
21/11 | Suspendido | ||||
28/11 | Defensa | B. Marenco | |||
05/12 | Deep Tempest | F. La Rocca | |||
12/12 | When the order matters: architectures for sequences | Octavia Camps (Northeastern) | |||
19/12 | TBA | Juan Cervino (MIT) |
ORGANIZADORES:
Si quiere dar una charla en el Seminario por favor escribir a los organizadores. Abajo hay algunas referencias con temas y artículos de posible interés (pero charlas de un tema distinto, en un área afin a la descripción del seminario estan bienvenidas!)
Referencias de interes:
Titulo | Tema | Link | EsClave |
Compressive Sampling | Compressive Sensing | [Link] | * |
Compressive Fourier | Compressive Sensing | [arXiv] | |
CS on measures | Compressive Sensing | [arXiv] | |
Rank Minimization (RM) via nuclear norm regularization | Matrix Factorization | [arXiv] | * |
Online Matrix Factorization | Matrix Factorization | [arXiv] | |
Big data is Low rank | Matrix Factorization | [arXiv] | |
Geometry of regularization | Matrix Factorization | [arXiv] | * |
Latent variable model selection | Matrix Factorization | [arXiv] | |
Wasserstein and Regularization in ML | DRO | [arXiv] | * |
Regularized Risk in DL | DRO | [arXiv] | |
Implicit Regularization Towards Rank Minimization in ReLU Networks | Implicit regularization | [arXiv] | |
Limitations of Implicit Bias in Matrix Sensing. | Implicit regularization | [arXiv] | |
Implicit Regularization in Deep Matrix Factorization | Implicit regularization | [arXiv] | |
Implicit Regularization in Tensor Factorization | Implicit regularization | [arXiv] | |
Matrix factorization geodesic convexity | Matrix Factorization | [arXiv] | |
Introducción | Geometría de Deep Learning | [arXiv] | |
Aproximación con neural networks | Geometría de Deep Learning | [arXiv], [arXiv] | |
Espacios de aproximación | Geometría de Deep Learning | [arXiv] | |
(FOCM) Theory-to-Practice Gap en DL | Geometría de Deep Learning | [arXiv] | |
Universality of transformers | DRO | [arXiv] | |
Conformal Prediction | [notas] | ||
Deep K-SVD Denoising | [arXiv] |