'LabelBench: A Comprehensive Framework for Benchmarking Adaptive Label-Efficient Learning'

by Jifan Zhang, Yifang Chen, Gregory Canal, Arnav Mohanty Das, Gantavya Bhatt, Stephen Mussmann, Yinglun Zhu, Jeff Bilmes, Simon Shaolei Du, Kevin Jamieson, Robert D Nowak

New paper on label-efficient supervised finetuning of LLMs.

We address the expensive prompt annotation cost by humans/proprietary LLMs, saving as much as 50% on FLAN V2.

Paper: arxiv.org/abs/2401.06692
Work led by: Jifan Zhang Yifang Chen Gantavya Bhatt Arnav Das
1/

Wondering how the SOTA poison removal technique behaves when the pre-training objective is changed? Join us today at 11 AM (~ 1 hour), the workshop of Backdoors in ML #NeurIPS2023 for our work - “Effective Backdoor Mitigation Depends on Pre-training Objectives”

📄Happy to introduce our recent work at TMLR - 'Accelerating Batch Active Learning Using Continual Learning Techniques'. We approach one of the major bottlenecks in Active Learning (AL): the necessity of training models from scratch every time. [1/N]

Accelerating Batch Active Learning Using Continual Learning Techniques

Arnav Mohanty Das, Gantavya Bhatt, Megh Manoj Bhalerao, Vianne R. Gao, Rui Yang, Jeff Bilmes.

Action editor: Changjian Shui.

openreview.net/forum?id=T55dL…

#forgetting #continual #retrained

Active Learning studies how AI best direct humans to provide feedback and annotation. Yet these algorithms have wildly different performances on new datasets, making it impossible to pick the best algo to use in practice. We now have a solution (appearing at NeurIPS 2023) 1/7

Our recent work on 'Accelerating Batch Active Learning Using Continual Learning Techniques' will appear in #ICML2023 workshop on 'Data-centric Machine Learning Research'. Joint work with Arnav Das and Jeff Bilmes. Summary 🧵 soon!

We present Epiphany, a neural network that predicts cell-type specific Hi-C contact maps of 3D genome folding using 5 commonly generated 1D epigenomic marks. Epiphany could also be used to study the contribution of specific epigenomic signals to 3D architecture.