Chen Xing Email & Phone Number

Palo Alto, Ca, Us

San Francisco, California, Us

Research focus:1. Up-to-date and challenging evaluation benchmarksWe keep evaluating the most frontier AI foundation models with hard evaluations as they evolve. We keep developing up-to-date evaluation benchmarks to find AI models’ weak points and contribute such benchmarks to the community.2. LLM’s self-evolution capabilities with/without humans in the loop.We keep investigating how well current AI foundation models can self-evolve. Whether a LLM can self-improve by generating data to train itself, is still a controversial question without concrete answers. Papers present evidence to both standing sides of this question and observations along this line are crucial to both Scale and the community. In this direction, we investigate: a. what are the best ways in practise to train models to self-evolve in different data domains, b. whether adding human supervision in the loop can significantly help, c. and most importantly, if human supervision can help, what are the most efficient ways to add human supervision.3. Frontier dataIn frontier data domains, such as data for agentic applications of LLMs, brainstorm and experiment with our customers from the top labs around the world on what would be the most efficient/effective way to generate data to make frontier models climb.

San Francisco, California, Us

Research focusing on the data side of LLM training, instruction tuning, use-case incubation and evaluation1. FOFO: A Benchmark to Evaluate LLMs’ Format-Following Capability (ACL 2024): https://arxiv.org/pdf/2402.18667.pdf2. Key contributor of Lemur, a 70B AI foundation model for AI agents (ICLR 2024 Spotlight): https://openreview.net/forum?id=hNhwSmtXRh& 3. Key contributor of X-Gen (Salesforce's first LLM trained in house from scratch): https://arxiv.org/abs/2309.03450

San Francisco, California, Us

1. A series of multi-modality (vision-language) research work: Open Vocabulary Object Detection with Pseudo Bounding-Box Labels (ECCV 2022) ULIP: Learning Unified Representation of Language, Image and Point Cloud for 3D Understanding (CVPR 2023) GlueGen: Plug and Play Multi-modal Encoders for X-to-image Generation (ICCV 2023) ......2. Incubated NLP/AI powered new features for Einstein Conversation Mining

Mountain View, Ca, Us

Trustworthy Deep Learning:Distance-Based Learning from Errors for Confidence Calibration (ICLR 2020)

Montreal, Quebec, Ca

Adaptive Cross-Modal Few-Shot Learning (NeurIPS 2019)

Montréal, Québec, Ca

Research Intern at Montréal Institute of Learning Algorithms (MILA) supervised by Prof. Yoshua BengioRepresentative work on analysis of deep learning optimization: A Walk with SGD (https://arxiv.org/abs/1802.08770)

Wa , Wa , Us

Research Intern at Natural Language Computing Team of Microsoft Research Asia supervised by Dr.Wei-Ying Ma

Redmond, Washington, Us

Research Intern at Machine Learning Group, supervised by Dr. Zheng Chen

Doctor Of Philosophy (Phd), Computer Science, Deep Learning

Bachelor'S Degree, Computer Science