@melissapan @mertcemri @shulynnliu @shagarw21 @AlexKrentsel @profjoeyg @matei_zaharia @istoica05 @AlexGDimakis Heartiest congratulations, Melissa!! 🎉
English
Deepanjali Mishra
33 posts
Deepanjali Mishra
@DeepCache
Ph.D. Candidate in Computer Architecture @CarnegieMellon | Rethinking data center architectures for efficiency 🌱. (she/her)
Pittsburgh, Pennsylvania Katılım Ocak 2023
7 Takip Edilen89 Takipçiler
Thrill to share that we received the best paper award for the FAGEN workshop at ICML 2026 🚀
It felt surreal walking through the poster session and seeing so much exciting work on agent failures. Where “failure taxonomy” become a standard terminology in the field, and MAST being one of the shared, systematic ways to organize negative examples. 🙏
Back in 2023, the MAST team had many discussions about how to share this kind of insight with the community and having the failure taxonomy as the main contribution, when such papers were still uncommon at top ML conferences. Now seeing an entire subcommunity form around agent failures is incredibly exciting.🔥
Despite agents being deployed everywhere, I still believe we’re only at the beginning. MAST captured the prevalent failure modes we observed at the time, but as agents expand into more diverse domains, we need more task-specific failure understanding to build better systems. Stay tuned for our full "MAST v2" paper! 🥁
And huge thanks to the FAGEN organizers (@wzenus) for bringing this community together, and to all the reviewers! 🫡
Alex Dimakis@AlexGDimakis
Best Paper Award to our ICML Workshop paper 'Fantastic Adaptive Taxonomies and How to Use Them'. Let me summarize briefly what the paper is about: Failure taxonomies are becoming increasingly important. We show that failure taxonomies can be used in multiple ways: 1. As a test-time scaling tool for best-of-N judges, 2) as a mutation feedback mechanism in optimization loops and 3) as runtime feedback for coding agents. Previously people used MAST and other hand-made fixed failure taxonomies. In this paper we show how to create the failure taxonomy adaptively and dynamically: We observe agent rollouts and create an adaptive failure taxonomy, bespoke to the agent weaknesses and task challenges. These adaptive taxonomies give a massive boost: On Terminal Bench 2 we get 89.9% with Opus 4.6 / Forgecode harness and the adaptive taxonomy used with a Best-of-N Judge, outperforming fixed taxonomies by 15%. (1/n)
English
🚨 NEW SINGLE JUST DROPPED 🚨
My new song “Bounded Autonomy” is out NOW on all platforms! 🔥🎶 Go stream it, save it, share it.
#NewMusic #NewSingle
Google Gemini@GeminiApp
We want to see what you’ve been making with Lyria 3 Pro in Gemini. 🎶 Share your creations in the replies 👇
English
English
Got interviewed for the first time at #NeurIPS while presenting our MAST poster 😳
Thanks to @J_Novikova_NLP for the interview and for having me in @WiAIR_podcast
Check out the video here: youtu.be/n9mM3GPaDXQ?si…
YouTube
English
@melissapan @lisabdunlap @shulynnliu @aczhu1326 @tsunghan_wu OMGGG, looking very cool, Melissssa 💛💛💛
English
The Sky’s Fun Committee, representing the ppl of sky, just dropped the new lab theme:
⚫️💖 Black Pink x Halloween 🎃🦇
We have:
- Gru & the minions
- kpop ???
🫰😉
English
@_emliu @gneubig @XiongChenyan @AdtRaghunathan @jacobandreas @AndrewLampinen Congratulations, Emmy!!!
English
MIA from twitter for a while, but brief life update!
✨ Passed my thesis proposal during the summer, a huge thank you to all my committee members! @gneubig, @XiongChenyan, @AdtRaghunathan, @jacobandreas, and @AndrewLampinen
✨ I will be at #EMNLP2025 next week (more l8r)
English
🎉 Excited to share a new chapter!
I'm thrilled to announce that I'll be joining Max Planck Institute for Software Systems (MPI-SWS) as a tenure-track faculty member! I'm grateful to my mentors, colleagues, collaborators, friends, and everyone who has supported me along the way.
English
Just bleached + dyed my hair for the first time at home, fully guided by Gemini 2.5 Pro and Perplexity (auto) 😬 turned out the models are surprisingly good at this “unverifiable” + manual task!!
English
@satishs I am doing well -- learning a lot! How have you been? 🙂
English
55 seconds faster in the 5K, in my 55th year. Old dog, new PB - turns out the legs still remember. Great pacing support, cool weather & loud cheers made this a fun morning. #Gratitude
English
@tianyin_xu @Jinghao_J @siebelschool Hi Tianyin, I am doing well—just wrapped up the first year of my PhD! It has been a great learning experience.
How have you been?
English
English
Yay! DOCTOR @Jinghao_J from @siebelschool !
Truly outstanding work and brilliant person!
Thank you, Jinghao, for showing me the characters of a pure OS hacker, builder, and researcher! Such a privilege and I'm so extremely grateful!
English
I will be presenting MAST at #MLSys2025 YPS Workshop poster session tomorrow. Come talk to me if you are into agent, or need project ideas for upcoming ddls - I have 14 of them ready for you 😉
Melissa Pan@melissapan
🚨 Why Do Multi-Agent LLM Systems Fail? ⁉️ 🔥 Introducing MAST: The first multi-agent failure taxonomy - consists of 14 failure modes and 3 categories, generalizes for diverse multi-agent systems and tasks! Paper: arxiv.org/pdf/2503.13657 Code: github.com/multi-agent-sy… 🧵1/n
English
@melissapan @Ponnapalli95 Such a wonderful initiative! I wonder what "creative coding" means here 🙂
English
Running “Creative Coding” workshops for grade5-8 girls at EYH today - co-design the course with @Ponnapalli95
English
Paper #4. HYDRA was a project @CSDatCMU on building computer systems for "AI" in early 70s. The machine (C.mmp) consists of 16 PDP-11 minicomputers connected to 32 MB of shared memory through a crossbar switch. HYDRA is the OS kernel for C.mmp. It's known for the design of "capability on steroids." The design is fascinating, despite not looking too practical retrospectively.
Tianyin Xu@tianyin_xu
In preparation of CS 523, I'm reading through many classic OS papers (just did Dijkstra's "THE" OS paper and Hansen's Nucleus paper). I plan to post my notes, as a forcing function to read papers carefully before going to class (inspired by @qianl_cs and @petereliaskraft). Bear with me if my posts are too boring or pedantic.
English
"As of 2024, the machine is on display at CMU, in Wean Hall, on the ninth floor." (en.wikipedia.org/wiki/C.mmp)
Sadly I didn't get the chance to check out the C.mmp machine last semester when I visited CMU. Next time!
English
English
Congratulations to Onur Mutlu on receiving the IEEE Computer Society 2025 Harry H. Goode Memorial Award for “seminal contributions to computer architecture research and practice, especially in memory systems“.
safari.ethz.ch/onur-mutlu-rec…
@_onurmutlu_ @IEEEAwards @ComputerSociety
English
Our work on debugging with GitHub Copilot Chat won Best Paper Award at VL/HCC! We introduce ROBIN, an “investigate & respond” agentic LLM workflow. As we approach a balance b/w automation vs. collaboration, we observe 3.5x greater bug fixes! 🤖
📜Paper: microsoft.com/en-us/research…
VL/HCC@vlhcc
🏆 Best Research Paper Award! 🏆 Huge congratulations to @imYbajpai and the team, from Microsoft for their award-winning paper: "Let's Fix This Together - Conventional Debugging with GitHub Copilot." 🤖💡 Well-deserved recognition! 🎉 #VLHCC2024 #BestPaperAward
English
I am super honoured to be selected as one of the 🌟2025 Siebel Scholars 🌟. I'm also on the job market this year for Applied Research positions. Please reach out if you think I would be a good fit.
siebelscholars.com/articles/siebe…
@dmguiuc @uiuc_nlp
English
English
Thrilled to share that I have started my PhD at @UCBerkeley with Prof. Aditya Parameswaran @adityagp!
My research will focus on using human-centered approaches to assist data scientists, developers, and end-users with their data needs. Excited for the journey ahead!
English
Honored to be selected as a 2024 MLSys Rising Star⭐!
Looking forward to meeting everyone at the workshop next month in Santa Clara!
#mlsys #risingstar #workshop #nvidia #machinelearning #ai #systems
MLCommons@MLCommons
Congratulations to the 2024 @MLCommons rising star cohort. Take a look at these impressive future leaders in Systems and ML research. mlcommons.org/2024/06/2024-m… #machinelearning #risingstars #systems #data
English