iiinova — where your digital self keeps evolving

Rahul G. Krishnan @rahulgk.bsky.social · 07/07/2026

If you're at ICML 2026, check out Valentyn's work on correcting posterior estimations from causal foundation models!

Rahul G. Krishnan @rahulgk.bsky.social · 29/11/2025

We're hiring tenure-stream faculty at all levels. web.cs.toronto.edu/employment-o... If you'd like to learn more about what being faculty here is like, please do reach out!

web.cs.toronto.edu

Employment Opportunities — Department of Computer Science, University of Toronto

Are you looking for a thought-provoking and inventive career at a leading institution?

120

Rahul G. Krishnan @rahulgk.bsky.social · 29/11/2025

Would love to connect with folks interested in automated reliable decision across industries! Finally, if you're on the job market this year. Join us at the University of Toronto Department of Computer Science in Canada.

120

Rahul G. Krishnan @rahulgk.bsky.social · 29/11/2025

Applying to graduate school next year? We're hiring! GPUs, coffee, an incredible city and mental space to do blue sky research @uoftcompsci.bsky.social @vectorinstitute.ai @uoftmedicine.bsky.social cs.toronto.edu/~rahulgk/lin...

cs.toronto.edu

121

Rahul G. Krishnan @rahulgk.bsky.social · 29/11/2025

(iii) D3M: a hypothesis test to leverage computation to detect deterioration of predictive models. @teivng.bsky.social cs.toronto.edu/~viet/d3m.html

cs.toronto.edu

100

Rahul G. Krishnan @rahulgk.bsky.social · 29/11/2025

(ii) MDM-Prime; a new class of discrete diffusion models that operates over subtokens. chen-hao-chao.github.io/mdm-prime/

chen-hao-chao.github.io

Beyond Masked and Unmasked: Discrete Diffusion Models via Partial Masking

MDM-Prime

100

Rahul G. Krishnan @rahulgk.bsky.social · 29/11/2025

My students and I will be at #NeurIPS2025 and EurIPS from Dec 2-8. My students and l will be presenting three papers.

100

Reposted by Rahul G. Krishnan

vahidbalazadeh.bsky.social @vahidbalazadeh.bsky.social · 11/06/2025

🚨 Introducing CausalPFN, a foundation model trained on simulated data for in-context causal effect estimation, based on prior-fitted networks (PFNs). Joint work with Hamid Kamkari, Layer6AI & @rahulgk.bsky.social 🧵[1/7] 📝 arxiv.org/abs/2506.07918 🔗 github.com/vdblm/Causal... 🗣️Oral@ICML SIM workshop

141

Rahul G. Krishnan @rahulgk.bsky.social · 23/04/2025

Theres lots more to do to understand CFT better, and build on it to create better post-training methods to fine-tune large language models. Reach out to me or Ethan if you're interested in collaborating on this or pushing this idea to new domains and problems!

010

Rahul G. Krishnan @rahulgk.bsky.social · 23/04/2025

📖 We’ve also open-sourced OpenMedText, integrating 121K biomedical articles & 29 medical textbooks to push future research in domain-adaptive fine-tuning in biomedicine.

110

Rahul G. Krishnan @rahulgk.bsky.social · 23/04/2025

🔧 We "negative" and "adaptive" prompts, confirming that the semantic content of prompts changes and impacts fine-tuning effectiveness.

100

Rahul G. Krishnan @rahulgk.bsky.social · 23/04/2025

📊 Results: On medical benchmarks, CFT improves accuracy by ~2.25% over CPT; in finance, it boosts performance by ~4.32%! Importantly, these gains scale effectively with larger models. 📈 Check out Appendix E.1 for preliminary results on GEMINI Flash 1.5M!

100

Rahul G. Krishnan @rahulgk.bsky.social · 23/04/2025

🏥 We tested this idea in biomedical (using newly curated OpenMedText dataset of journals & textbooks!) and financial data—CFT significantly outperforms continued pretraining (CPT) and instruction fine-tuning (IFT) in zero-shot settings.

100

Rahul G. Krishnan @rahulgk.bsky.social · 23/04/2025

🎓 Instead of using Q&A as in instruction tuning, CFT uses reflective instructions (e.g., "Reflect on how what you will see changes what you know...") motivated by how humans learn.

100

Rahul G. Krishnan @rahulgk.bsky.social · 23/04/2025

💡Contextual finetuning (CFT) uses contextual prompts during fine-tuning to adaptively change the semantic understanding that LLMs leverage during the process of learning new information.

100

Rahul G. Krishnan @rahulgk.bsky.social · 23/04/2025

🚀 Problem: Language models struggle with rapidly evolving info and context in fields like medicine & finance. We need ways to teach LLMs new information and control how they absorb this knowledge. 🔍 Insight: Why not explain and teach LLMs how to learn?

100

Rahul G. Krishnan @rahulgk.bsky.social · 23/04/2025

My student, Ethan Choi, will be at #ICLR2025 presenting Contextual Finetuning (CFT) and teaching LLMs how to learn (joint work with Muhammad Adil Asif, Ziwen Han, John Willes @vectorinstitute.ai) 🌟Project page: younwoochoi.github.io/cft-iclr/ #239, April 26 10-12:30(Hall3,2B)

younwoochoi.github.io

120

Rahul G. Krishnan @rahulgk.bsky.social · 21/04/2025

If it helps, I usually learn something new (either directly or from further digging) about the behavior of markets.

210

Reposted by Rahul G. Krishnan

UofT-TCAIREM @uoft-tcairem.bsky.social · 24/02/2025

📣T-CAIREM member @rahulgk.bsky.social's presentation is online! From Associational to Causal Predictions with #DeepLearning: An examination of recent advances in bridging the gap between associative #neuralnetworks and causal reasoning. 🎥 www.youtube.com/watch?v=yE6S...

youtube.com

Rahul G. Krishnan | From associational to causal predictions with deep learning

YouTube video by Schwartz Reisman Institute

012

Rahul G. Krishnan @rahulgk.bsky.social · 28/01/2025

Come by tomorrow to hear about what we have been up to!

120

Rahul G. Krishnan @rahulgk.bsky.social · 21/12/2024

I thought about this a bit, I think helping PhD students close the translational gap from research to deployment (in industry or their own startups), particularly if they don't want to go into academia, is one way forward.

040

Rahul G. Krishnan @rahulgk.bsky.social · 21/12/2024

o3 is incredible! Since we've maxed out scale and $$$ on scaling inference-time compute I hope we now get back to thinking about the right combination of neural nets and algorithm to performant models cheaper, faster, and more reliably.

011

Reposted by Rahul G. Krishnan

Aviraj (Avi) Newatia @projectavi.bsky.social · 14/12/2024

1/6 Presenting "Unlearning Tabular Data without a 'Forget Set'"! We explore a new unlearning algorithm RELOAD in tabular learning. Drop by @neuripsconf.bsky.social Workshop on Table Representation Learning (@trl-research.bsky.social): - SAT 14 Dec from 2:30pm-3:15pm! - East Meeting Room 11-12

511

Rahul G. Krishnan @rahulgk.bsky.social · 12/12/2024

Are you around at Neurips? Would love to say hi and catch up!

110

Rahul G. Krishnan @rahulgk.bsky.social · 12/12/2024

Come by our poster today to learn about decision making under unobserved confounding!

110

Rahul G. Krishnan @rahulgk.bsky.social · 11/12/2024

Finally, if you're interested in understanding how to leverage energy-based normalizing flows, check out Lance's work on Meow (chienfeng-hub.github.io/meow/) He'll be presenting on Dec. 12, 11:00 AM–2:00 PM at West Ballroom A-D #6403 🧵(7/7)

chienfeng-hub.github.io

Maximum Entropy Reinforcement Learning via Energy-Based Normalizing Flow

100

Rahul G. Krishnan @rahulgk.bsky.social · 11/12/2024

@nikitadhawan.bsky.social developed NATURAL (www.cs.toronto.edu/~nikita/natu...) with @cottascience.bsky.social , Karen & @cmaddis.bsky.social. Its an end-to-end pipeline that starts from raw-text data and ends with a causal (**) effect associated with an intervention. (**) conditions apply 🧵(6/7)

cs.toronto.edu

NATURAL

151

Rahul G. Krishnan @rahulgk.bsky.social · 11/12/2024

b] ~Billions of dollars each year are spent on trials to assess interventions. Can we use crowdsourced data to know which intervention is likely to work ahead of time? Doing so requires answering a causal question! But the data to answer this question is locked in unstructured text. 🧵(5/7)

101

Rahul G. Krishnan @rahulgk.bsky.social · 11/12/2024

Find Vahid to learn more about in-context causal inference and lots of other cool problems that he spends his time thinking about! 🧵(4/7)

120

Rahul G. Krishnan @rahulgk.bsky.social · 11/12/2024

In arxiv.org/abs/2404.07266, Vahid shows how to use offline expert data with unobserved confounding to guide decision making using a nonparametric prior to guide learning policies for bandits, MDPs, and POMDPs. Thu 12 Dec 4:30 - 7:30 pm PST 📷 West Ballroom A-D Poster #6708 🧵(3/7)

arxiv.org

Sequential Decision Making with Expert Demonstrations under Unobserved Heterogeneity

We study the problem of online sequential decision-making given auxiliary demonstrations from experts who made their decisions based on unobserved contextual information. These demonstrations can be v...

131

Rahul G. Krishnan @rahulgk.bsky.social · 11/12/2024

a] Today, we learn from data and treat it as ground truth -- should we? A doctor often knows more about their patient than is represented in electronic medical records. A teacher knows more about their students than what their grades suggest. 🧵(2/7)

110

Rahul G. Krishnan @rahulgk.bsky.social · 11/12/2024

First post! I’ll be at @NeurIPSConf #NeurIPS2024 until Sunday. I'd love to chat about causality for medicine & science. I'm also looking for a postdoc interested in experimental design for medicine, if that's you, send me a message. I'll be presenting two papers at the main conference. 🧵(1/7)

120