Navid Azizan

arxiv.org HardFlow: Hard-Constrained Sampling for Flow-Matching Models via... Diffusion and flow-matching have emerged as powerful methodologies for generative modeling, with remarkable success in capturing complex data distributions and enabling flexible guidance at...

MIT LIDS @MITLIDS · Sep 16

When #AI operates in the physical world, “pretty close” may not cut it. LIDS researchers developed HardFlow, a method that helps pretrained generative AI satisfy hard constraints while improving solution quality — without retraining. bit.ly/4xydxBN

View quoted post on X

Young-Jin Park @young_j_park · Jun 19, 2025

🧠 Inference-time scaling lets LLMs spend more compute to solve harder problems, but not every question needs that!

After all, we don’t use a whiteboard to solve 1 + 1. So why should an LLM?

Introducing Instance-Adaptive Inference-Time Scaling, a smarter way to allocate compute during inference.

💡 By calibrating PRMs, we can estimate how likely each reasoning step is to lead to a correct answer. This lets us:
- Use less compute on easy questions and on likely incorrect reasoning paths
- Spend more compute on harder questions and on promising reasoning steps
🎯 Same accuracy, up to 50% less compute.

Paper → arxiv.org/abs/2506.09338
Code → github.com/azizanlab/inst…
Calibration dataset → huggingface.co/datasets/young…
Calibrated PRMs → huggingface.co/collections/yo…

Joint work with @KGreenewald, @KavehAlim, @HW_HaoWang, and @NavidAzizan.

View quoted post on X

arxiv.org Training-Free Uncertainty Estimation for Embedding Models Embedding models, often obtained via self-supervised learning, extract general-purpose representations from data. Quantifying the reliability of these representations is crucial, as many...

MIT LIDS @MITLIDS · Jul 19, 2024

How to assess a general-purpose AI model’s reliability before it’s deployed. A new technique from MIT LIDS researchers @NavidAzizan and Young-Jin Park enables users to compare several large models and choose the one that works best for their task. bit.ly/3zNWiEf

View quoted post on X