Zhifei Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Zhifei Li has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- Ion Stoica
University of California, Berkeley, Berkeley, USA
2 shared publications - Akshat Gupta
University of California - Berkeley, Berkeley, USA
1 shared publication - Alexander Krentsel
University of California - Berkeley, Berkeley, USA
1 shared publication - Alexandros Georgios Dimakis
University of California, Berkeley, Berkeley, USA
1 shared publication - Armando Fox
University of California, Berkeley, Berkeley, USA
1 shared publication - Ashwin Naren
University of California - Berkeley, Berkeley, USA
1 shared publication
Latest Publications
- Open Access
- 26 May 2026
- Published by Association for Computing Machinery (ACM)
- p. 1223-1227
SkyDiscover: A Flexible, Adaptive Framework for AI-Driven Scientific and Algorithmic Discovery
LLM-driven evolutionary search is emerging as a powerful approach for discovering algorithms and designs, but existing frameworks are difficult to reuse, extend, and compare. We present SkyDiscover, a flexible, adaptive framework for AI-driven scientific and algorithmic discovery. SkyDiscover decomposes the discovery loop into four reusable components: Context Builder, Solution Generator, Evaluator, and Solution Selector, while exposing the control logic above them as a programmable interface. This modular design enables rapid experimentation and even supports adaptive designs where AI can adapt or even optimize the optimization process itself during search. We demonstrate SkyDiscover across more than 200 optimization tasks spanning mathematical optimization, systems design, algorithmic programming, and constrained image generation. Under fixed budgets and shared models, the adaptive algorithms implemented on top of SkyDiscover achieves the strongest open-source performance compared to OpenEvolve, ShinkaEvolve, and GEPA, and matches or exceeds AlphaEvolve on many tasks. A live demo further showcases end-to-end discovery with real-time monitoring, human-in-the-loop steering, and meta-optimization of the search process.
- 26 May 2026
- Published by Association for Computing Machinery (ACM)
- p. 1223-1227
Open AccessSkyDiscover: A Flexible, Adaptive Framework for AI-Driven Scientific and Algorithmic Discovery
LLM-driven evolutionary search is emerging as a powerful approach for discovering algorithms and designs, but existing frameworks are difficult to reuse, extend, and compare. We present SkyDiscover, a flexible, adaptive framework for AI-driven scientific and algorithmic discovery. SkyDiscover decomposes the discovery loop into four reusable components: Context Builder, Solution Generator, Evaluator, and Solution Selector, while exposing the control logic above them as a programmable interface. This modular design enables rapid experimentation and even supports adaptive designs where AI can adapt or even optimize the optimization process itself during search. We demonstrate SkyDiscover across more than 200 optimization tasks spanning mathematical optimization, systems design, algorithmic programming, and constrained image generation. Under fixed budgets and shared models, the adaptive algorithms implemented on top of SkyDiscover achieves the strongest open-source performance compared to OpenEvolve, ShinkaEvolve, and GEPA, and matches or exceeds AlphaEvolve on many tasks. A live demo further showcases end-to-end discovery with real-time monitoring, human-in-the-loop steering, and meta-optimization of the search process.
- Open Access
- 26 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1349-1364
SkyWalker: A Locality-Aware Cross-Region Load Balancer for LLM Inference
- 26 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1349-1364
Open AccessSkyWalker: A Locality-Aware Cross-Region Load Balancer for LLM Inference
- Open Access
- 18 February 2025
- Published by Association for Computing Machinery (ACM)
- p. 1385-1386
Checkpoint: A Tool for Supporting Terminal-Based Capture-the-Flag Assessments
Checkpoint is a responsive, terminal-based assignment system designed to support faculty in teaching and assessing a broad scope of computer science topics with Capture-the-Flag (CTF) style challenges. Unlike existing CTF frameworks, where binary completion of a given task is the only metric for evaluation, Checkpoint enables students to receive partial credit and dynamic feedback as they complete tasks. Not only does this allow students to confirm they are on the right track, but it simultaneously provides detailed analytics to instructors about each student's progress as they complete the assignment. Presented by an in-browser terminal emulator neatly wrapped in a Docker image, Checkpoint is a versatile tool that can be implemented in a variety of Learning Management Systems (LMSs) with minimal effort. Instructors can easily define flags in a user-friendly configuration file with flag names mapped to optional feedback and regex patterns, which automatically mark flags as complete and render feedback to students within their terminal environment.
- 18 February 2025
- Published by Association for Computing Machinery (ACM)
- p. 1385-1386
Open AccessCheckpoint: A Tool for Supporting Terminal-Based Capture-the-Flag Assessments
Checkpoint is a responsive, terminal-based assignment system designed to support faculty in teaching and assessing a broad scope of computer science topics with Capture-the-Flag (CTF) style challenges. Unlike existing CTF frameworks, where binary completion of a given task is the only metric for evaluation, Checkpoint enables students to receive partial credit and dynamic feedback as they complete tasks. Not only does this allow students to confirm they are on the right track, but it simultaneously provides detailed analytics to instructors about each student's progress as they complete the assignment. Presented by an in-browser terminal emulator neatly wrapped in a Docker image, Checkpoint is a versatile tool that can be implemented in a variety of Learning Management Systems (LMSs) with minimal effort. Instructors can easily define flags in a user-friendly configuration file with flag names mapped to optional feedback and regex patterns, which automatically mark flags as complete and render feedback to students within their terminal environment.