Past Research Deepika Bablani

Learned Step Size Quantization (LSQ)

International Conference on Learning Representations (ICLR)

A method for learning the quantization step size end-to-end during training, enabling low-precision neural network inference with minimal accuracy degradation.

Quantization Efficient inference ICLR

Neural Inference at the Frontier of Energy, Space and Time

Science

An exploration of the fundamental limits of neural network inference across the axes of energy consumption, memory footprint, and computational time with implications for hardware-aware model design.

Efficient inference Science

SiLQ: Simple Large Language Model Quantization-Aware Training

ACL Findings

A simple and scalable quantization-aware training approach for large language models, enabling efficient deployment without significant loss in downstream task performance.

LLM quantization ACL

Entropy Approximation Guided Layer Selection (EAGL) for Mixed-Precision Neural Network Quantization

ASPLOS '24 Workshop on Energy Efficient Machine Learning and Cognitive Computing

A principled approach to selecting which layers to quantize and at what precision, guided by an entropy-based sensitivity approximation reducing the search cost for mixed-precision configurations.

Mixed precision Quantization ASPLOS

Improving Transfer Using Augmented Feedback in Progressive Neural Networks

NeurIPS Workshop on Cognitively Informed Artificial Intelligence

An investigation into how augmented feedback signals improve lateral knowledge transfer in progressive neural networks, drawing inspiration from cognitive science models of learning.

Transfer learning NeurIPS Cognitively inspired

Incorporating Attention in World Models for Improved Dynamics Modeling

NeurIPS Workshop on Modeling the Physical World

An extension of world model architectures with attention mechanisms, improving the accuracy and generalization of learned dynamics models for physical prediction tasks.

World models Attention NeurIPS