Talks

I will give a brief review of how large-language models are now being used for theoretical physics research. I will show the rapid progress of these models at the example of the TPBench benchmark, and present our recent work on improving their reliability with a symbolic verification agent and test-time scaling techniques. I will also discuss whether these models are truly reasoning and speculate how we might improve their performance in our field in the future.

An unavoidable part of studying astrophysics based on catalogs of detected events is quantifying the probability of detecting different types of events. I will briefly discuss the types of design considerations that go into constructing such estimates and how they will scale with larger catalog sizes. I will also introduce the wide variety of uses for such data products, including uncovering unexpected features within the data caused by the fact that humans build and operate the detectors.

Binary neutron star mergers are critical for understanding the dynamics of dense matter, the origin of gravitational waves, and the formation channels of the heaviest elements through the r-process. I will review how long-lived remnants can act as central engines for multimessenger observations. I will then discuss how we can identify phase transitions within neutron stars or their remnants using such observations. Phase transitions alter the system’s dynamics and can produce distinct observable signatures, potentially detectable with next-generation facilities and observatories. These signatures can be used to probe matter at supranuclear densities and to test fundamental physics.

Heavy right-handed neutrinos are highly motivated due to their connection with the origin of neutrino masses via the seesaw mechanism. If the right-handed neutrino Majorana mass is at or below the weak scale, direct experimental discovery of these states is possible in laboratory experiments. However, there is no a priori basis to expect right-handed neutrinos to be so light since the Majorana mass is a technically natural parameter and could comfortably reside at any scale, including at scales far above the weak scale. Here we explore the possibility that the right-handed neutrino Majorana mass originates from electroweak symmetry breaking. Working within an effective theory with two Higgs doublets, nonzero lepton number is assigned to the bilinear operator built from the two Higgs fields, which is then coupled to the right-handed neutrino mass operator. In tandem with the neutrino Yukawa coupling, following electroweak symmetry breaking a seesaw mechanism operates, generating the light SM neutrino masses along with right-handed neutrinos with masses below the electroweak scale. This scenario leads to novel phenomenology in the Higgs sector, which may be probed at the LHC and at future colliders. There are also interesting prospects for neutrinoless double beta decay and lepton flavor violation. We also explore some theoretical aspects of the scenario, including the technical naturalness of the effective field theory and ultraviolet completions of the right-handed neutrino Majorana mass.

The Dark Energy Spectroscopic Instrument (DESI) is the first of a new generation of Dark Energy experiments and probes evolution in the universe using galaxy clustering. Within the galaxy clustering signal, the projected location of the Baryon Acoustic Oscillations (BAO) acts as a standard ruler to map cosmic evolution. I will present the latest BAO results from the DESI Data Release 2 (DR2) sample, which contains 3 years of data, and their impact on our understanding of dark energy and neutrino masses. Finally, I will consider how the amplitude of the BAO signal can help us measure the Hubble constant, potentially helping to solve the Hubble tension.

We consider the problem of finding the optimal value of n in the n-step temporal difference (TD) learning algorithm. Our objective function for the optimization problem is the average root mean squared error (RMSE). We find the optimal n by resorting to a model-free optimization technique involving a one-simulation simultaneous perturbation stochastic approximation (SPSA) based procedure. Whereas SPSA is a zeroth-order continuous optimization procedure, we adapt it to the discrete optimization setting by using a random projection operator. We prove the asymptotic convergence of the recursion by showing that the sequence of n-updates obtained using zeroth-order stochastic gradient search converges almost surely to an internally chain transitive invariant set of an associated differential inclusion. This results in convergence of the discrete parameter sequence to the optimal n in n-step TD. Through experiments, we show that the optimal value of n is achieved with our algorithm for arbitrary initial values. We further show using numerical evaluations that the proposed algorithm outperforms a well known discrete parameter stochastic optimization algorithm ‘Optimal Computing Budget Allocation (OCBA)’ on benchmark RL tasks. 

When a neural network becomes extremely wide or deep, its learning dynamics simplify and can be described by the same “mean-field” ideas that explain magnetism and fluids. I will walk through these ideas step-by-step, showing how they suggest practical recipes for initialization and optimization that scale smoothly from small models to cutting-edge transformers. I will also discuss neural scaling laws—empirical power-law rules that relate model size, data, and compute—and illustrate them with solvable toy models.