Eugene Belilovsky

Associate Professor@Concordia University·加拿大·OpenReview

研究方向

Learned Optimizers · Federated Learning · Continual Learning · Visual Relation · Computer Vision · CNN · Cross-Modal Representations · Generative Models · Structured Prediction · Sparsity and Structured Sparsity

Eugene Belilovsky

$\mu$LO: Compute-Efficient Meta-Generalization of Learned Optimizers

Celo2: Towards Learned Optimization Free Lunch

Less is More: Undertraining Experts Improves Model Upcycling

Model Parallelism With Subnetwork Data Parallelism

Communication Efficient LLM Pre-Training with SparseLoCo

Towards Efficient Unroll Generalization in Learned Optimizers

PETRA: Parallel End-to-end Training with Reversible Architectures

Controlling Forgetting with Test-Time Data in Continual Learning

ACCO: Accumulate While You Communicate for Communication-Overlapped Sharded LLM Training

AdaFisher: Adaptive Second Order Optimization via Fisher Information

Accelerating Training with Neuron Interaction and Nowcasting Networks

$\mu$LO: Compute-Efficient Meta-Generalization of Learned Optimizers

ACCO: Accumulate while you Communicate, Hiding Communications in Distributed LLM Training

WASH: Train your Ensemble with Communication-Efficient Weight Shuffling, then Average

From Feature Visualization to Visual Circuits: Effect of Model Perturbation

Investigating Grokking phenomena below the Critical Data Regime

FairDropout: Using Example-Tied Dropout to Enhance Generalization for Minority Groups