
Assistant professor @UCLAComSci. Better ML via better data, Machine learning, Optimization
Los Angeles, CA
Joined July 2018
- Tweets78
- Following308
- Followers1.6K
- Likes276
Pinned Tweet
Weโre thrilled by the amazing response to our #ICML2024 tutorial on โFoundations of data-efficient learningโ! Over 1000 attendees joined us. Thank you all! ๐๐ฑ๐ฑ๐ฑ
โก๏ธ Slides: baharanm.github.io/ICML24_tuโฆ
โก๏ธ Recording: will be available on Aug 22 ๐๐
I'll be giving a 2-hour tutorial on data-efficient learning with my PhD student @sjoshi804 on Monday July 22 at #ICML2024. Join us to learn more about this cool topic! โก๏ธ We can learn better from better data! โฌ
๏ธ๐๐ฑ
Baharan Mirzasoleiman retweeted
Announcing the ๐๐๐๐ฌ๐ญ ๐๐๐ฉ๐๐ซ ๐๐ฐ๐๐ซ๐๐ฌ๐ for DATA-FM workshop @iclr_conf: ๐ถ๐๐๐ ๐ฟ๐๐ ๐ ๐ก๐ ๐น๐๐ก ๐๐๐๐: ๐๐๐๐๐๐๐๐ ๐ท๐๐ก๐ ๐๐๐ข๐๐๐๐ ๐ผ๐๐๐๐๐ฃ๐๐ ๐๐๐๐๐๐๐ง๐๐ก๐๐๐ ๐๐ ๐น๐๐๐ก๐ by @jiayuan_ye_ et al. arxiv.org/abs/2604.08519 (morning oral session)
๐ฅ๐๐จ๐ง๐จ๐ซ๐๐๐ฅ๐ ๐๐๐ง๐ญ๐ข๐จ๐ง goes to ๐ด๐๐โ๐๐๐๐ ๐ท๐๐๐๐๐๐๐: ๐๐๐๐ฃ๐๐๐๐ฆ ๐
๐๐๐ข๐๐๐๐ ๐ถ๐๐๐ฆ๐๐๐โ๐ก ๐
๐๐ ๐ ๐๐๐ ๐ด๐๐ฆ ๐ฟ๐๐๐๐ข๐๐๐ ๐๐๐๐๐ by @jcqln_h et al. arxiv.org/abs/2602.07120 (afternoon oral session)
Congratulations to all the authors, and please join us at the workshop tomorrow at ๐๐จ๐จ๐ฆ ๐๐๐ ๐+๐ for more discussions around everything data! ๐ Updated schedule at data-fm-iclr2026.github.io/, with invited talks by @sewon__min @baharanm @jcniebles @fredsala!
Thank you to our sponsors @SnorkelAI and @datologyai for all your support!
Can weak LLMs supervise strong LLMs to obtain superior performance? ๐ค Yes!! ๐คฉ
Which weak models are better supervisors? ๐ค
Check out @xue_yihao65785โs awesome #icml2025 paper to know how to identify best weak supervisors without having to collect labels! ๐๐ฑ
๐ Our paper โRepresentations Shape Weak-to-Strong Generalizationโ is accepted at #ICML2025!
We study weak-to-strong generalization (W2SG)โa core problem in superalignmentโand offer new insights into the role of models' internal representations in W2SG.
1/
Baharan Mirzasoleiman retweeted
๐จ Join us at the Workshop on Spurious Correlation & Shortcut Learning (SCSL) at #ICLR2025!
@iclr_conf
๐๏ธ April 28, 2025
๐ Garnet 214-215, Singapore EXPO
๐ More info: scslworkshop.github.io
#ICLR2025
Want to train LLMs with better performance and lower GPU memory requirements on data mixtures?
Check out this cool #ICLR2025 paper of @dangnth97 this Friday April 25 at 10am, Poster #265
๐๐ฑ
๐ Achievement unlocked: having papers with all of my labmates and somehow all ended up at ICLR!
Iโll be presenting our work โMini-batch Coresets for Memory-efficient Language Model Training on Data Mixturesโ at #ICLR2025 ๐ธ๐ฌ
Come by and chat! ๐ on Fri, Apr 25 | 10 AM GMT+8
1
19
1,972
Can we pretrain deep models with small synthetic data?
Dataset Distillation via Knowledge Distillation is the way to go!
Check out @sjoshi804โs #ICLR2025 paper this Saturday April 26 at 9am, Poster #307 ๐๐ฑ
#ICLR2025
Can you pre-train deep models with small, synthetic datasets? ๐คฏ
We introduce the first effective dataset distillation method for self-supervised learning (SSL) โ boosting downstream accuracy by up to 13% over baselines.
๐งช Poster #307, Sat Apr 26, 9am
Big congrats @YuYang_i on your graduation!! ๐๐ ๐
very nice PhD thesis with great contributions ๐ฑ
Iโm proud of all youโve done, and I wish you the best! ๐
Sharing a little late update (before itโs no longer news): I wrapped up my PhD at the end of last year and recently joined @OpenAIโs reasoning team ๐โจ!
56
6,008
Baharan Mirzasoleiman retweeted
(2/2) Not at UCLA but interested in this work? Check arxiv.org/abs/2502.02407. Thanks to our fantastic intern @unregularized (soon to return full-time!) for leading this project, along with collaborators @ynd and Atish. Thanks to my UCLA host @baharanm for the seminar invitation!
Baharan Mirzasoleiman retweeted
We are delighted that our proposal for the Workshop on โSpurious Correlation and Shortcut Learning: Foundations and Solutionsโ has been accepted at @iclr_conf 2025, hosting many brilliant keynote speakers and panelists. Stay tuned:
scslworkshop.github.io
@SCSLWorkshop 1/
Baharan Mirzasoleiman retweeted
At NeurIPS? Check out the 2nd workshop on Attributing Model Behavior at Scale (ATTRIB)!
Meeting Rm 205-207, starting @ 9am - amazing talks from @SurbhiGoel_ @sanmikoyejo @baharanm, Robert Geirhos, and @coallaoh + exciting contributed talks!
More info: attrib-workshop.cc/
Iโll help presenting our #NeurIPS2024 posters tomorrow (Friday):๐ฑ
1- Changing the training data distribution to improve in-distribution performance (11@west #7106) w. @dangnth97
2- Data selection for fine-tuning LLMs with superior performance (16:30@west #5401) w. @YUYANG_UCLA
Baharan Mirzasoleiman retweeted
Attending NeurIPS'24?
Please mark your calendar for our special event "SFU@NeurIPS 2024" sites.google.com/view/sfu-atโฆ
9 speakers from both academia & industry!
Only a 10-min walk from the convention center!
Letโs enjoy exciting talks and open discussions!
Same training and test distribution yields optimal in-distribution performance?
@dangnth97 showed in his #NeurIPS2024 paper that this is not true when training with gradient methods!!๐ฎ๐
Changing the training data distribution yields SOTA!๐
Check it out Fri Dec 13, 11am, PS#5
Smaller high-quality subsets of language data not only improve LLMsโ training efficiency, but also yield considerably better performance! ๐๐๐ฑ
@YUYANG_UCLA has a theoretically-rigorous method for this in her #NeurIPS2024 paper!
Check it out on Fri, Dec 13, 16:30, #PS 6
1/ I'll be at #NeurIPS2024 presenting our work SmallToLarge (S2L): Data-efficient Fine-tuning of LLMs! ๐
Whatโs S2L? Itโs a scalable data selection method that trains a small proxy model to guide fine-tuning for larger models, reducing costs while preserving performance. ๐
Baharan Mirzasoleiman retweeted
Assist. Prof. Baharan Mirzasoleiman @baharanm of @UCLAComSci & her large-scale machine learning research group @UCLA is part of the new @NSF-@SimonsFdn Institute for Cosmic Origins at @UTAustin that aims to use AI to research the mysteries of the cosmos. cns.utexas.edu/news/announceโฆ
Baharan Mirzasoleiman retweeted
๐ข @UCLAComSci is hiring! Open to all CS areas!
- Multiple Tenure-track Assistant Professor Positions: recruit.apo.ucla.edu/JPF0979โฆ
- Open Rank Teaching Professor Position: recruit.apo.ucla.edu/JPF0980โฆ
(We hired 11 Assistant Professors in the past two years ...)
Baharan Mirzasoleiman retweeted
The Adversarial Machine Learning Rising Star Awards deadline is in two weeks! Submit your application and help us promote your work and research vision! @trustworthy_ml @LLMSecurity @ml_safety @safe_paper
๐ฉ(1/2)
Please help forward the Call for the 2024 Adversarial Machine Learning (AdvML) Rising Star Awards!
We promote junior researchers in AI safety, robustness, and security. Award events are hosted at AdvML'Frontiers workshop @NeurIPSConf 2024
Info: sites.google.com/view/advml/โฆ
Iโll also present โSafeClipโ on behalf of @WenhanYang0315 tomorrow at 1:30pm (poster session 6) #814. See you there! ๐
CLIP is highly sensitive to data poisoning and backdoor attacks. In this #ICML2024 paper, @WenhanYang0315 proposed an interesting way to pretrain CLIP robust to such attacks without compromising the performance! ๐ฑ๐ฑ
๐arxiv.org/pdf/2310.05862
Thu, July 25, Poster session 6, #814
Iโll present โMixProโ on behalf of @xue_yihao65785 tomorrow at 11:30 (poster session 5) poster #800. Come check it out ๐
ML models are sensitive to distribution shift. Can we adapt a model with only a few examples from the target domain? In this #ICML2024 paper, @xue_yihao65785 proposes an effective way, with nice theoretical analysis๐ฑ
๐arxiv.org/pdf/2305.14521
Thu, July 25, Poster session 5, #800


























