The loss landscape of transformer models trained on protein sequences exhibits Lipschitzian properties analogous to those studied in parameterized split feasibility problems, enabling stability certificates for learned representations.
Adversarial Debate Score
17% survival rate under critique
Model Critiques
Supporting Research Papers
- Performative Scenario Optimization
This paper introduces a performative scenario optimization framework for decision-dependent chance-constrained problems. Unlike classical stochastic optimization, we account for the feedback loop wher...
- ParetoEnsembles.jl: A Julia Package for Multiobjective Parameter Estimation Using Pareto Optimal Ensemble Techniques
Mathematical models of natural and man-made systems often have many adjustable parameters that must be estimated from multiple, potentially conflicting datasets. Rather than reporting a single best-fi...
- On Lipschitzian properties of multifunctions defined implicitly by"split"feasibility problems
In the present paper, a systematic study is made of quantitative semicontinuity (a.k.a. Lipschitzian) properties of certain multifunctions, which are defined as a solution map associated to a family o...
- Sampling at intermediate temperatures is optimal for training large language models in protein structure prediction
We investigate the parameter space of transformer models trained on protein sequence data using a statistical mechanics framework, sampling the loss landscape at varying temperatures by Langevin dynam...
Formal Verification
Z3 checks whether the hypothesis is internally consistent, not whether it is empirically true.