Code implementing the experiments of "Fixes That Fail Self-Defeating Improvements in Machine-Learning Systems".
最近更新: 接近3年前Study of a hypothetical attack scenario where an attacker corrupts images that are fed into an AI system, and has only a few chances before risking being detected.
最近更新: 接近3年前We introduce the notion of a social catalyst in the context of social media a person who prompts interaction between their friends through their posts. We find not only that this activity is heavily skewed, but that it also correlates with participation in groups and events. We validate the behavioral measure using a survey wherein people nominate social catalysts in their friend network. Whether a person will be nominated as a social catalyst can be predicted using network features and the tendency of their posts to prompt discussion. Although individuals' role as social catalysts has not attracted much study to date, this paper demonstrates that it can be studied both via behavioral and survey measures.
最近更新: 接近3年前This code reproduces the results of the paper, "Measuring Data Leakage in Machine-Learning Models with Fisher Information"
最近更新: 接近3年前DiffQ performs differentiable quantization using pseudo quantization noise. It can automatically tune the number of bits used per weight or group of weights, in order to achieve a given trade-off between model size and accuracy.
最近更新: 接近3年前This repository provides code for "On Interaction Between Augmentations and Corruptions in Natural Corruption Robustness".
最近更新: 接近3年前Experiment code to create the dataset presented in the EACL2021 paper "FEWS: Large-Scale, Low-Shot Word Sense Disambiguation with the Dictionary".
最近更新: 接近3年前Starter code for SoundSpaces challenge at CVPR 21's Embodied AI workshop
最近更新: 接近3年前PyTorch code for ICLR 2021 paper Unbiased Teacher for Semi-Supervised Object Detection
最近更新: 接近3年前
