arXiv subjects

Dennis Duan

Publications and source records attributed to Dennis Duan.

3 recordsLinked to original sources

arXiv2025

Restructuring Vector Quantization with the Rotation Trick

Vector Quantized Variational AutoEncoders (VQ-VAEs) are designed to compress a continuous input to a discrete latent space and reconstruct it with minimal distortion. They operate by maintaining a set of vectors -- often referred to as the codebook -- and quantizing each encoder output to the nearest vector in the codebook. However, as vector quantization is non-differentiable, the gradient to the encoder flows around the vector quantization layer rather than through it in a straight-through approximation. This approximation may be undesirable as all information from the vector quantization operation is lost. In this work, we propose a way to propagate gradients through the vector quantization layer of VQ-VAEs. We smoothly transform each encoder output into its corresponding codebook vector via a rotation and rescaling linear transformation that is treated as a constant during backpropagation. As a result, the relative magnitude and angle between encoder output and codebook vector becomes encoded into the gradient as it propagates through the vector quantization layer and back to the encoder. Across 11 different VQ-VAE training paradigms, we find this restructuring improves reconstruction metrics, codebook utilization, and quantization error. Our code is available at https://github.com/cfifty/rotation_trick.

cs.LG↗

arXiv2024

Context-Aware Meta-Learning

Large Language Models like ChatGPT demonstrate a remarkable capacity to learn new concepts during inference without any fine-tuning. However, visual models trained to detect new objects during inference have been unable to replicate this ability, and instead either perform poorly or require meta-training and/or fine-tuning on similar objects. In this work, we propose a meta-learning algorithm that emulates Large Language Models by learning new visual concepts during inference without fine-tuning. Our approach leverages a frozen pre-trained feature extractor, and analogous to in-context learning, recasts visual meta-learning as sequence modeling over datapoints with known labels and a test datapoint with an unknown label. On 8 out of 11 meta-learning benchmarks, our approach -- without meta-training or fine-tuning -- exceeds or matches the state-of-the-art algorithm, P>M>F, which is meta-trained on these benchmarks. Our code is available at https://github.com/cfifty/CAML.

cs.LG↗

arXiv2022

Carbide: Highly Reliable Networks Through Real-Time Multiple Control Plane Composition

Achieving highly reliable networks is essential for network operators to ensure proper packet delivery in the event of software errors or hardware failures. Networks must ensure reachability and routing correctness, such as subnet isolation and waypoint traversal. Existing work in network verification relies on centralized computation at the cost of fault tolerance, while other approaches either build an over-engineered, complex control plane, or compose multiple control planes without providing any guarantee on correctness. This paper presents Carbide, a novel system to achieve high reliability in networks through distributed verification and multiple control plane composition. The core of Carbide is a simple, generic, efficient distributed verification framework that transforms a generic network verification problem to a reachability verification problem on a directed acyclic graph (DAG), and solves the latter via an efficient distributed verification protocol (DV-protocol). Equipped with verification results, Carbide allows the systematic composition of multiple control planes and realization of operator-specified consistency. Carbide is fully implemented. Extensive experiments show that (1) Carbide reduces downtime by 43% over the most reliable individual underlying control plane, while enforcing correctness requirements on all traffic; and (2) by systematically decomposing computation to devices and pruning unnecessary messaging between devices during verification, Carbide scales to a production data center network.

cs.NI↗