Krithik Ramesh (@KrithikTweets) / X

Krithik Ramesh

2,693 posts

Krithik Ramesh

@KrithikTweets

AI + Math @MIT, compbio stuff @broadinstitute, prev: research @togethercompute

Joined February 2017

Pinned
Krithik Ramesh
@KrithikTweets
Mar 21, 2025
🧬 Meet Lyra, a new paradigm for accessible, powerful modeling of biological sequences. Lyra is a lightweight SSM achieving SOTA performance across DNA, RNA, and protein tasks—yet up to 120,000x smaller than foundation models (ESM, Evo). Bonus: you can train it on your Mac. read
122K
Krithik Ramesh
@KrithikTweets
Dec 12, 2021
Holy crap,unbelievable #WTF1 #Formula1 #MaxVerstappen33
Krithik Ramesh
@KrithikTweets
Aug 21, 2025
I just wanted to give a quick shoutout to @PrimeIntellect (specifically @afurgs and @willccbb) for reaching out so thoughtfully and not killing my instance when I forgot to setup auto top off :)
5.8K
Krithik Ramesh
@KrithikTweets
Jan 26, 2025
Replying to @willccbb
Djikstra’s algorithm dropped 2 days ago and people can’t stop getting creative with it. 10 wild examples a 🧵
922
Krithik Ramesh
@KrithikTweets
Jan 28, 2025
One of my favorite classes was Geometric Methods for Machine Learning with the incredible @mweber_PU! The biggest shift in my mind after taking her class is thinking about DL architecture as a means of preserving certain properties (permutation equivariance, homophily, etc.)
Petar Veličković
@PetarV_93
Jan 28, 2025
And so it begins -- year 4 of the Geometric DL course @Cambridge_Uni! It feels surreal every time 😁 Thank you @charlieharris01 @chaitjo for the photos 🙌
11K
Krithik Ramesh
@KrithikTweets
Sep 3, 2025
Replying to @ScrewderiaF1
Unfortunately it’s gonna be one DNF and one putting leclerc on hards when it’s pouring.
2K
Krithik Ramesh
@KrithikTweets
Mar 21, 2025
Replying to @KrithikTweets
Breaking down how biological sequences encode molecular functions remains a central challenge in computational biology. For example, given a GFP sequence, can we predict its fluorescence brightness? To unify biological sequence modeling across DNA, RNA, and proteins into a
3.6K
Krithik Ramesh
@KrithikTweets
Mar 21, 2025
Replying to @KrithikTweets
We drew a mathematical connection between State Space Models (SSMs) and polynomial approximation, showing how their hidden states can naturally approximate the polynomial terms that govern epistatic relationships. This makes SSMs ideal for modeling biological functions as
2.7K
Krithik Ramesh
@KrithikTweets
Mar 9, 2025
This is genuinely one of the best blogposts about subquadratic primitives/ linear attention I’ve read. @SonglinYang4 is incredible at providing clear intuition motivating architectural decisions. Worth the read! sustcsonglin.github.io/blog/2024/delt…
2K
Krithik Ramesh
@KrithikTweets
Mar 25, 2025
I got interested in subquadratic primitives around when Based dropped—Zoology, Based, JRT were some of the best papers I’ve ever read. They made me fall in love with this work. Learned so much about thinking across architectures, hardware, and theory from the brilliant folks at
Simran Arora
@simran_s_arora
Mar 25, 2025
BASED ✌️ turns 1! One year since its launch at NeurIPS 2023 — and it's helped shape the new wave of efficient LMs. ⚡️ Fastest linear attention kernels 🧠 405B models trained on 16 GPUs 💥 Inspired Mamba-v2, RWKVs, MiniMax Checkout our retrospective below!
2.1K
Krithik Ramesh
@KrithikTweets
Jan 26, 2025
Replying to @willccbb
This is a throwback to a memory I simply forgot I possessed
1.8K
Krithik Ramesh
@KrithikTweets
Mar 21, 2025
Replying to @KrithikTweets
On a personal note I want to give a huge shoutout to @Sameed for being such a great friend and mentor. It's been a privilege of a lifetime getting to work with @_albertgu, Michael Mitzenmacher, and @PardisSabeti. Excited to see how this can advance biology and bring machine
1.8K
Krithik Ramesh
@KrithikTweets
Mar 21, 2025
Replying to @KrithikTweets
Lyra achieves state-of-the-art results in 6 out of 7 intrinsically disordered protein region tasks, with an average AUC of 0.89, outperforming the ProtT5-based model (average AUC of 0.86). Remarkably, Lyra accomplishes this using only 55K parameters, compared to ProtT5’s 3
1.7K
Krithik Ramesh
@KrithikTweets
Mar 21, 2025
Replying to @KrithikTweets
Check out our paper "Lyra: An Efficient and Expressive Subquadratic Architecture for Modeling Biological Sequences," detailing how mathematical insights overcome computational limitations in biology. arxiv.org/abs/2503.16351 A joint work with @Sameed @_albertgu Michael
arxiv.org
Lyra: An Efficient and Expressive Subquadratic Architecture for...
Deep learning architectures such as convolutional neural networks and Transformers have revolutionized biological sequence modeling, with recent advances driven by scaling up foundation and...
2K