Eli Weinstein
eliweinstein.bsky.social
Eli Weinstein
@eliweinstein.bsky.social
Incoming assistant professor of chemistry at the Technical University of Denmark (DTU). Also at Jura Bio. machine learning, statistics, chemistry, biophysics


https://eweinstein.github.io/
We demonstrate empirically and prove theoretically that LeaVS can dramatically accelerate learning, increasing the effective dataset size by orders of magnitude.
October 21, 2025 at 2:38 PM
Crucially, it depends on jointly modifying the experimental protocol and the training algorithm: on their own, neither modification helps.
October 21, 2025 at 2:38 PM
This approach lets you focus limited measurements on the most informative datapoints, maximizing information gain without compromising reliability.
October 21, 2025 at 2:38 PM
Second, modify the training algorithm: compensate for the missing negatives by incorporating the generative variational synthesis model into the objective.
October 21, 2025 at 2:38 PM
First, modify the experiment: only measure positive examples of functional proteins. Don't spend a limited sequencing budget on any negatives.
October 21, 2025 at 2:38 PM
In this paper we describe a method to overcome this measurement bottleneck.
October 21, 2025 at 2:38 PM
To test, we can deliver billions of designs to different cells. But there is a cost to recovering those designs' function, to obtain (x,y) data.
October 21, 2025 at 2:38 PM
With variational synthesis, we can now build quadrillions of generative model-designed sequences. The bottleneck is now testing, not synthesis.
October 21, 2025 at 2:38 PM
Scaling up protein ML requires understanding and eliminating bottlenecks in the design-build-test-learn cycle.
October 21, 2025 at 2:38 PM
If you are interested in working with me as a student or postdoc, or otherwise collaborating, please reach out.
May 29, 2025 at 5:04 PM