PhD Candidate in Computer Science @Harvard · Co-organizer @capp_lac · Previously: @quantilma @uniandes · Colombiana 🇨🇴
Cambridge, MA
Joined July 2009
- Tweets4.4K
- Following1.2K
- Followers1K
- Likes4K
Data selection is just as critical as the training itself. But how do you choose which samples to use?
Our new work explores what works and what doesn't when selecting a subset of instructions to tune on. Awesome thread by @nihalcanrun below! 👇
Targeted instruction tuning for LLMs involves selecting a subset of instructions from a candidate pool using a small query set from target tasks. Despite growing interest, we still lack guidance on what to select. Our new preprint brings clarity to this space (thread 👇).
4
257
Paula Rodríguez Díaz retweeted
AI has transformed system design across the computing stack.
We've seen AI agents that can optimize compilers and generate high-performance code, test-to-text regression that can model complex production systems, and reinforcement learning agents that automatically solve chip placement and circuit synthesis.
In our new paper, we step back and ask: what actually generalizes across these domains?
Paula Rodríguez Díaz retweeted
📄 New preprint alert:
We study 🪃Boomerang Distillation🪃, a surprising phenomenon that allows generating a family of pre-trained LLMs of intermediate sizes from a single teacher–student pair — 𝐧𝐨 𝐞𝐱𝐭𝐫𝐚 𝐭𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐫𝐞𝐪𝐮𝐢𝐫𝐞𝐝!
🧵👇
Excited to present our paper at #UAI2025 on Thursday 🇧🇷
Can we predict which datasets transfer best when downstream decisions, not just predictions, matter? Can we know this before learning?
We introduce OTD³, a new dataset distance for PtO tasks
arxiv.org/pdf/2409.06997
Paula Rodríguez Díaz retweeted
I'm hiring for a machine learning data scientist & research assistant for summer 2025!
Join me on a project on invasive species management with an innovative startup doing on-the-ground removal of environmentally destructive invasives.
Paid, full-time w/ possibility to extend.
ALT Join me in working on a project on invasive species management, by predicting species risk and planning capture strategies. This work will be in partnership with an innovative startup that is working directly to capture environmentally destructive invasive species in the US. This project will be to build predictive machine learning models and design harvest strategies for Inversa’s invasive species removal efforts. The primary goal is to develop usable ML models and optimization tools to inform practical environmental decision-making on the ground. There will also be opportunities to extend this work into a research publication for a top-tier AI venue. How to apply: Please send a CV and 3–5 paragraphs of your background/interests via email to <[email protected]> with the subject line "Application: ML for invasive species management". Applications will be reviewed on a rolling basis.
Paula Rodríguez Díaz retweeted
¡Hola! El #SeminarioCAPP de marzo tiene como invitada a Emily Black (NYU) @em__black, presentando su trabajo sobre (In)Justicia en Sistemas de IA: Auditorías Tributarias y Multiplicidad de Modelos🧵.
🗓 Jueves, 6 de marzo.
🕦11:30 a. m. COL.
💻Zoom: tinyurl.com/SeminarioCAPP
Paula Rodríguez Díaz retweeted
(1/9) Excited to share my recent work on "Alignment reduces LM's conceptual diversity" with @TomerUllman and @jennhu, to appear at #NAACL2025! 🐟
We want models that match our values...but could this hurt their diversity of thought?
Preprint: arxiv.org/abs/2411.04427
Paula Rodríguez Díaz retweeted
📢 ¡Este jueves 6 de febrero regresa nuestro seminario mensual en colaboración con @QuantilMA!
💡 Además, únete a nuestra comunidad de investigadores en IA, computación y política pública. Conéctate, comparte y colabora en nuestro canal de Slack: tinyurl.com/rvjsd662
Paula Rodríguez Díaz retweeted
These measures will slightly impact U.S. consumers but will devastate entire industries in Colombia.
The American public should know that Colombia’s government is using this as a distraction from a growing internal humanitarian crisis caused by its erratic security policies.
Paula Rodríguez Díaz retweeted
I am recruiting PhD students to join my lab at Harvard in Fall 2025! (deadline Dec 15)
If you are interested in solving problems at the intersection of reinforcement learning, imitation learning, and NLP, pls consider applying (bit.ly/4fnficx)! @hseas @KempnerInst
Paula Rodríguez Díaz retweeted
📢 We are hiring a postdoctoral fellow in ML and Environmental Data Science! 📈🌏 This interdisciplinary position is co-hosted by the CS department 💻 and the Environmental Data Science Innovation and Inclusion Lab (ESIIL) 🌱 at CU Boulder 🦬. jobs.colorado.edu/jobs/JobDe…
Paula Rodríguez Díaz retweeted
Please retweet: I am recruiting PhD students at Berkeley!
Please apply to @Berkeley_EECS or @UCJointCPH if you are interested in ML applied to health, inequality, or social science, and mention my name in your app.
More details on work/how to apply: cs.cornell.edu/~emmapierson/
Paula Rodríguez Díaz retweeted
🚀 <24 hours to go until #EAAMO2024 kicks off! We can't wait to welcome you to @LaUASLP tomorrow, whether you'll be joining us in person or virtually.
Conference information can be found at - conference.eaamo.org/confere…
Looking forward to 3 days of insightful collaboration! 💡
Paula Rodríguez Díaz retweeted
📢 I’m recruiting PhD students for Fall'25 in @Berkeley_EECS or @UCJointCPH! Please apply if you have related interests in AI for social good / networks / public health / social science
See my website for more on my research and how to apply: serinachang5.github.io/
Please RT!
Paula Rodríguez Díaz retweeted
Interested in becoming a postdoctoral fellow at the
@Harvard Center for Research on Computation & Society @HCRCS? Now soliciting applications for Academic year 2025-26.
Deadline: Dec 2
crcs.seas.harvard.edu/apply
Paula Rodríguez Díaz retweeted
A reminder that second colloquium in our series focused on Latin America and the Caribbean is starting in less than three hours!
Date: Wednesday, September 25, 2024. Time: 12:00 - 1:00 PM ET.
Join us via this Zoom link: tinyurl.com/eaamozoom
We propose a *decision-aware dataset distance* that goes beyond traditional metrics. Using OT and factoring in the *impact of decisions*, our approach (right) better measures similarity in PtO tasks and improves transferability prediction across domains with distribution shifts
1
5
469
We empirically demonstrate the effectiveness of our method across three PtO tasks: topk selection, shortest path optimization, and inventory management. Results show that incorporating decisions into OT-based dataset distances consistently improves PtO transferability prediction
1
3
372
We presented preliminary versions of this work at the @DMLRWorkshop and the Humans, Algorithmic Decision-Making and Society workshops at ICML 2024. We're excited to announce that a more detailed version is now available on arXiv: arxiv.org/abs/2409.06997
7
341
