Perimeter Institute for Theoretical Physics

Deep networks are often thought of as black boxes. Their ability to encompass vast swathes of knowledge indeed makes them hard to explain. Yet many of their behaviours — generalization under overparametrization, grokking, OOD failures, neural scaling laws — recur across architectures and scales, and each, however surprising, can be reproduced and explained in controlled settings. I will review these efforts to identify and explain the universal phenomena of deep learning, and suggest that an LLM may amount to a sum of such tractable sub-phenomena, interpolative in nature. Finally, leaving scientific rigor aside, I'll argue that what separates this prosaic picture from the apparent magic of LLMs may well be the industrial scale of compute and human labour behind it, and that AGI in its deeper extrapolative sense may be much further away than claimed.

Sisi Zhou examines how quantum information methods can improve the precision of physical measurements. Qubit examples distinguish noise that quantum error correction can suppress from noise indistinguishable from the signal. The discussion extends to several qubits, phase estimation, many-body measurements and distributed sensors. A second part applies quantum-learning tools, including tomography, certification and purification, to estimation of several parameters. Because incompatible observables prevent all parameters from being measured optimally together, designing measurements is difficult. The talk presents practical protocols for several such sensing tasks that combine efficient implementation with sensitivity close to the optimum.

Uddipan Banik derives a dynamical account of dark-matter halo density profiles using the Boltzmann–Poisson equations for an expanding Universe populated by subhalos of many masses. A self-similar solution links density slope to halo mass growth; including gravitational scattering and subhalo capture gives gamma = 6/(s + 2). Initial collapse with s = 2 produces a prompt cusp of slope 3/2. Hierarchical growth connects the slope to the primordial power spectrum and recovers the Syer–White relation during matter domination. Stable clustering selects an inner Navarro–Frenk–White cusp of slope one, while stalled accretion approaches the outer slope of three. The framework connects these profiles to successive stages of halo assembly.

This talk is about the proposal that (1) quantum states that violate the momentum and Hamiltonian constraints represent instances of time, and (2) time evolution corresponds to the gradual collapse of an initial state toward a diffeomorphism-invariant state, driven by stochastic fluctuations of the lapse and shift. During the wavefunction collapse, the scale factor increases monotonically and thus acts as a clock. Scalar, vector, and tensor gravitons arise as physical excitations. In the long-time limit, the tensor gravitons exhibit emergent unitary dynamics, while the extra modes are damped by the non-unitary dynamics that suppress constraint-violating excitations. I will also comment on possible physical implications of these extra graviton modes.

Taillte May investigates how unstable black-hole quasinormal-mode spectra affect observed gravitational-wave signals. Small changes in a gravitational potential can substantially shift the frequencies normally used to test astrophysical black holes. A simple time-domain model reveals a delay before those shifts appear, tied to the echo travel time between the perturbation and the main potential. In the perturbative regime, the frequency change after one echo agrees with the conventional frequency-domain prediction. Beyond that regime, the intermediate waveform cannot be represented by a single constant frequency shift.

Arefe Abghari tests whether the reported excess in the quasar number-density dipole challenges the Cosmological Principle. If large-scale homogeneity and isotropy hold, the motion responsible for the cosmic microwave background dipole should also predict the dipole in distant-source counts. The analysis examines quasar selection and dipole-estimation systematics. Higher multipoles in the quasar density map, plausibly caused by survey and selection effects, can explain the apparent discrepancy. The talk argues that these effects need careful control before interpreting the measurements as evidence against the standard cosmological model.

Two-dimensional topological order has caught much attention in connection to topological phases of matter and topological quantum computation, in particular. It is known that tensor networks are useful tools to study mathematical structures appearing in these topics. We present recent understanding of them from a viewpoint of operator algebras. No knowledge on operator algebras are assumed.

One of the main achievements of my last four years at Perimeter has been the development and formulation of intrinsic null geometry. This framework has served as a common denominator for the study of the (algebraic) quantization of gravity on null hypersurfaces; connections to quantum-gravity phenomenology for causal diamonds and the derivation of the Verlinde-Zurek fluctuation identity; a complete characterization of the null gravitational phase space; links to asymptotic null infinity; and the related classification of eBMS anomalies. After introducing the basic toolkit of null geometry, I will digress on Carrollian connections and on how the intrinsic geometry is related to an ambient manifold. I will then introduce the null Brown-York stress tensor, which allows the intrinsic gravitational constraints to be recast as conservation laws. Finally, I will briefly retrace the path through the aforementioned applications, with particular emphasis on new and ongoing results.

Tree-level open string amplitudes are integrals over the moduli space of n points on the projective line, and their field theory limit is the Cachazo-He-Yuan formula for tr(\Phi^3) amplitudes. Joint work with Freddy Cachazo, Alfredo Guevara and Sebastián Mizera (CEGM) replaces the line by P^(k-1), giving amplitude-like objects whose Feynman diagrams are higher-dimensional polyhedral complexes rather than trees. After reviewing this framework, I will discuss recent progress on new factorization patterns which appear and linear zero-preserving deformations, which produce nonlinear sigma model amplitudes at k=2 by a result of Arkani-Hamed, Cao, Dong, Figueiredo and He, and new factorization behavior at k greater than or equal to 3.

Luca Mrini presents work in progress on quantising singular configuration spaces using the non-smooth calculus of metric-measure spaces. The framework generalises smooth quantum mechanics through Hilbert spaces, position and momentum operators, Hamiltonians, unitary dynamics, and observable algebras. A method for bounding mass gaps after symplectic reduction uses the curvature of the unreduced configuration space. Examples include the harmonic oscillator constrained to zero angular momentum, lattice Yang–Mills theory, and fractional-dimensional Laakso spaces. The talk closes with prospective applications to the Yang–Mills mass-gap problem and quantum gravity.

Kenneth Wharton of San Jose State University discusses local weak values in quantum circuits and whether they support a covariant description of individual qubits. The talk considers entanglement, constraints across an entire circuit and the relationship between quantum systems and classical spacetime. The confirmed Quantum Foundations seminar takes place on 10 September 2026, from 11:00 to 12:30 Toronto time in Perimeter Institute's Bob Room.

Jadwiga Wilkens of Johannes Kepler University Linz presents gate-set-shadow methods for characterising noise on a 36-qubit superconducting processor. The seminar examines reconstructed noise channels, correlated errors and cross-talk, including limitations of standard noise models. The confirmed Quantum Information seminar is scheduled for 9 September 2026, from 11:00 to 12:30 Toronto time in Perimeter Institute's Bob Room.

Zohar Ringel of the Hebrew University of Jerusalem presents a renormalisation-group approach to deep learning based on neural-network field theories. The seminar examines scaling laws, the removal of non-learnable field modes, benign overfitting and open questions about feature learning. The confirmed Quantum Matter seminar takes place in the Bob Room at Perimeter Institute on 8 September 2026, from 15:30 to 17:00 Toronto time.

Benoit Assi examines limits on reliable theoretical predictions for present and future particle colliders. Low-order simulations of QCD radiation carry large uncertainties, hadronization is commonly modeled rather than derived, and effective theories contain more operators than measurements can constrain. Information theory and machine learning can incorporate improved calculations into simulations, quantify uncertainty, choose informative observables and advance hadronization theory. Geometry of effective-theory field space combines infinite operator families into finite physical quantities, while information measures identify the combinations experiments can resolve. The talk develops these complementary approaches to extracting the available information from LHC and future-collider data.

Claudio Manzari connects astrophysical observations and precision flavor measurements to two unresolved aspects of CP symmetry. For the strong CP problem, the talk considers QCD axions produced in a supernova core and converted to gamma rays by surrounding magnetic fields. A Galactic supernova could expose this distinctive transient, motivating the GALAXY network of telescopes monitoring the full sky. The second part examines the origin of flavor-sector CP violation through precise measurements of the CKM unitarity triangle. Possible hints of spontaneous CP violation are discussed alongside their implications for quark flavor and the fundamental symmetries of matter.

Ania Liu examines gravitational-wave lensing as a probe of compact objects, dark matter and cosmic large-scale structure. Propagation over cosmological distances can distort signals, but identifying lensing in observations requires disentangling those distortions from uncertain waveform models and other astrophysical effects. The talk describes this phenomenology and the ambiguities it creates for interpretation, then presents recent methods and results from lensing searches using LIGO–Virgo–KAGRA data.

Arka Banerjee introduces k-nearest-neighbor distributions as summaries of cosmological survey data that capture information beyond two-point statistics. They respond to moments of all N-point correlations while retaining a computational cost comparable to two-point measurements. The talk covers auto-correlations and cross-correlations in discrete and continuous datasets, their relationship to other higher-order summaries, and applications that improve detection significance or parameter constraints. It also explores modeling these distributions with methods already successful for two-point functions in real and redshift space.

Tien-Tien Yu examines observational signatures of ultralight scalar dark matter with quadratic couplings to Standard Model fields. The couplings change fundamental constants over time, affecting primordial light-element production, the microwave-background power spectrum and tests of the equivalence principle. The talk follows these effects from the early Universe to present experiments and explains how current cosmological measurements constrain this class of scalar-dark-matter models.

Subha Majumdar presents a six-dimensional Milky Way phase-space catalogue designed to map stellar motions and dark matter beyond the reach of Gaia parallaxes alone. Gaia astrometry is combined with spectrophotometric distances and radial velocities from fourteen surveys, including DESI, SDSS-BOSS, APOGEE and LAMOST. The resulting catalogue contains about 33 million tracers and detailed distance–velocity measurements for roughly half a million halo stars. It characterizes distant clusters, dwarf galaxies and stellar streams, enabling mass modeling, kinematic studies, abundance mapping, Galactic archaeology and dark-matter searches. An application constructs a continuous Galactic rotation curve extending to 250 kiloparsecs.

Giada Santoro examines strong-field gravity and compact-object structure using GW250114, a gravitational-wave event with signal-to-noise ratio around eighty. The unusually strong signal supports precise tests of departures from general relativity and detailed measurements of black-hole ringdown. No tidal deformability is detected: the analysis places a ninety-percent upper bound of 34.8 on the effective tidal-deformability parameter. This is consistent with the vanishing tidal response predicted for Kerr black holes.