TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Supporting ML Reproducibility

2 点作者 dpleban超过 3 年前
TL;DR: We announced that we&#x27;d support the Reproducibility Challenge $500 per paper reproduced, and we&#x27;re announcing the award winners for the Spring &#x27;21 edition, as well as our support for the Fall &#x27;21 edition of the challenge. Check out the awesome papers below<p>---<p>Hey HN! Creator of DagsHub here. We really care about reproducibility. That&#x27;s why, a while back, we announced our support for the Papers with Code ML Reproducibility Challenge, and that we&#x27;d award participants $500 per paper reproduced (according to the guidelines), to align incentives and put our money where our mouth is!<p>Today, I&#x27;m really happy to share the teams that were given the award, and the projects they worked on – read the full blog here: https:&#x2F;&#x2F;dagshub.com&#x2F;blog&#x2F;ml-reproducibility-challenge-spring-2021&#x2F;<p>I honestly think the full read is interesting and worth your time, but here are the highlights from the papers:<p>1. Contextual Decomposition Explanation Penalization (CDEP) – The original paper proposes a method to reduce the chance of models learning spurious correlations instead of the actually important features. The team that reproduced it re-implemented the original project in Tensorflow, rewriting some functions completely from scratch! Along the way, they made a contribution to the Tensorflow addons repo<p>2. Self-supervision for Few-shot Learning – As its name suggests, this paper tests the importance of self-supervised learning in few-shot learning contexts. The team that reproduced it explored different input configurations than the one proposed in the article, and found out that it significantly affects the performance.<p>3. GANSpace: Discovering Interpretable GAN Controls – A proposed method to use &quot;simple&quot; PCA to create controls for GANs that are more humanly interpretable while being more computationally efficient. The team re-implemented the original implementation in Tensorflow and trained the model with a few benchmark datasets, they have a lot of very cool examples of the method in their report.<p>Thank you to everyone who took part in this challenge! None of this could be possible without you and we learned a lot in this process!<p>So what&#x27;s next – well we&#x27;ve decided to continue the support the Fall 2021 edition of the Reproducibility Challenge! We want to host more reproduced papers since this makes the ML field better for everyone.<p>If you want to take part and move the field forward on the reproducibility front, check out the guidelines for more information on how to take part: https:&#x2F;&#x2F;dagshub.com&#x2F;DAGsHub-Official&#x2F;reproducibility-challenge&#x2F;wiki&#x2F;ML+Reproducibility+Challenge+Fall+2021

暂无评论

暂无评论