Greedy SLIM

Greedy algorithms, characterized by their iterative, locally optimal choices, are being explored across diverse machine learning applications. Current research focuses on improving their efficiency and effectiveness in areas such as recommender systems (Greedy SLIM), large language model inference acceleration ("stairs" assisted greedy generation), and vision graph neural networks (GreedyViG), often demonstrating significant performance gains compared to alternative approaches. This focus on greedy methods stems from their computational advantages, particularly in high-dimensional or real-time settings, offering a valuable trade-off between computational cost and solution quality across various domains.

Papers

April 6, 2022

Greedier is Better: Selecting Multiple Neighbors per Iteration for Sparse Subspace Clustering
Jwo-Yuh Wu, Liang-Chi Huang, Wen-Hsuan Li, Chun-Hung Liu, Rung-Hung Gau
View Selection Neighborhood Selection Greedy SLIM Orthogonal Matching Pursuit Order Neighbor

February 10, 2022

Characterizing and overcoming the greedy nature of learning in multi-modal deep neural networks
Nan Wu, Stanisław Jastrzębski, Kyunghyun Cho, Krzysztof J. Geras
Return Conditioned Supervised Learning Greedy SLIM Greedy Learning

Greedy SLIM

Papers

Greedier is Better: Selecting Multiple Neighbors per Iteration for Sparse Subspace Clustering

Characterizing and overcoming the greedy nature of learning in multi-modal deep neural networks