Deep Learning
Speculative Decoding with Big Little Decoder
Figure 1: Illustration of (Left) the normal autoregressive decoding procedure of a large model and (Right) BiLD that consists of a small model and a large model. In BiLD, the small model generates tokens autoregressively (i.e., sequentially) until it hands over control to the large model.
Collaborative Alignment of NLP models
"concepts"--dictating desired model responses to certain inputs.
Equivariant Neural Simulators for Stochastic Spatiotemporal Dynamics
However, for many applications this formulation is insufficient.
Ego4D Goal-Step: Toward Hierarchical Understanding of Procedural Activities
It provides dense annotations for 48K procedural step segments (430 hours) and high-level goal annotations for 2,807 hours of Ego4D videos.
GEQ: Gaussian Kernel Inspired Equilibrium Models
Moreover, GEQ can be perceived as a weight-tied neural network with infinite width and depth. GEQ also enjoys better theoretical properties and improved overall performance.