Publications

Spike sorting is a crucial step in electrophysiological studies of neuronal activity. While many spike sorting packages are available, there is little consensus about which are most accurate under different experimental conditions. SpikeForest is an open-source and reproducible software suite that benchmarks the performance of automated spike sorting algorithms across an extensive, curated database of ground-truth electrophysiological recordings, displaying results interactively on a continuously-updating website. With contributions from eleven laboratories, our database currently comprises 650 recordings (1.3 TB total size) with around 35,000 ground-truth units. These data include paired intracellular/extracellular recordings and state-of-the-art simulated recordings. Ten of the most popular spike sorting codes are wrapped in a Python package and evaluated on a compute cluster using an automated pipeline. SpikeForest documents community progress in automated spike sorting, and guides neuroscientists to an optimal choice of sorter and parameters for a wide range of probes and brain regions.

During the first 2 hours of Drosophila development, precisely orchestrated nuclear cleavages, cytoskeletal rearrangements, and directed membrane growth lead to the formation of an epithelial sheet around the yolk. The newly formed epithelium remains relatively quiescent during the next hour as it is patterned by maternal inductive signals and zygotic gene products. We discovered that this mechanically quiet period is disrupted in embryos with high levels of dNTPs, which have been recently shown to cause abnormally fast nuclear cleavages and interfere with zygotic transcription. High levels of dNTPs are associated with robust onset of oscillatory two-dimensional flows during the third hour of development. Tissue cartography, particle image velocimetry, and dimensionality reduction techniques reveal that these oscillatory flows are low dimensional and are characterized by the presence of spiral vortices. We speculate that these aberrant flows emerge through an instability triggered by deregulated mechanical coupling between the nascent epithelium and three-dimensional yolk. These results highlight an unexplored connection between a core metabolic process and large-scale mechanics in a rapidly developing embryo.

Many problems in contemporary astrophysics---from understanding the formation of black holes to untangling the chemical evolution of galaxies---rely on knowledge about binary stars. This, in turn, depends on discovery and characterization of binary companions for large numbers of different kinds of stars in different chemical and dynamical environments. Current stellar spectroscopic surveys observe hundreds of thousands to millions of stars with (typically) few observational epochs, which allows binary discovery but makes orbital characterization challenging. We use a custom Monte Carlo sampler (The Joker) to perform discovery and characterization of binary systems through radial-velocities, in the regime of sparse, noisy, and poorly sampled multi-epoch data. We use it to generate posterior samplings in Keplerian parameters for 232,531 sources released in APOGEE Data Release 16. Our final catalog contains 19,635 high-confidence close-binary (P < few years, a < few AU) systems that show interesting relationships between binary occurrence rate and location in the color-magnitude diagram. We find notable faint companions at high masses (black-hole candidates), at low masses (substellar candidates), and at very close separations (mass-transfer candidates). We also use the posterior samplings in a (toy) hierarchical inference to measure the long-period binary-star eccentricity distribution. We release the full set of posterior samplings for the entire parent sample of 232,531 stars. This set of samplings involves no heuristic "discovery" threshold and therefore can be used for myriad statistical purposes, including hierarchical inferences about binary-star populations and sub-threshold searches.

The female meiotic spindles of most animals are acentrosomal and undergo drastic morphological changes while transitioning from metaphase to anaphase. The ultra-structure of acentrosomal spindles, and how this enables such dramatic rearrangements remains largely unknown. To address this, we applied light microscopy, large-scale electron tomography and mathematical modeling of female meiotic C. elegans spindles undergoing the transition from metaphase to anaphase. Combining these approaches, we find that meiotic spindles are dynamic arrays of short microtubules that turn over on second time scales. The results show that the transition from metaphase to anaphase correlates with an increase in the number of microtubules and a decrease of their average length. To understand the mechanisms that drive this transition, we developed a mathematical model for the microtubule length distribution that considers microtubule growth, catastrophe, and severing. Using Bayesian inference to compare model predictions and data, we find that microtubule turn-over is the major driver of the observed large-scale reorganizations. Our data suggest that cutting of microtubules occurs, but that most microtubules are not severed before undergoing catastrophe.

We study how to estimate a nearly low-rank Toeplitz covariance matrix T from compressed measurements. Recent work of Qiao and Pal addresses this problem by combining sparse rulers (sparse linear arrays) with frequency finding (sparse Fourier transform) algorithms applied to the Vandermonde decomposition of T. Analytical bounds on the sample complexity are shown, under the assumption of sufficiently large gaps between the frequencies in this decomposition. In this work, we introduce random ultra-sparse rulers and propose an improved approach based on these objects. Our random rulers effectively apply a random permutation to the frequencies in T's Vandermonde decomposition, letting us avoid frequency gap assumptions and leading to improved sample complexity bounds. In the special case when T is circulant, we theoretically analyze the performance of our method when combined with sparse Fourier transform algorithms based on random hashing. We also show experimentally that our ultra-sparse rulers give significantly more robust and sample efficient estimation then baseline methods.

Binary black holes (BBHs) detected by gravitational wave (GW) observations could be broadly divided into two formation channels: those formed through field binary evolution and those assembled dynamically in dense stellar systems. Each of these formation channels, and their sub-channels, populate a distinct region in the effective spin-mass (χeff−M) plane. Depending on the branching ratio of different channels, an ensemble of BBHs could show a trend in this plane. Here we fit a mass-dependent distribution for χeff to the GWTC-1 BBHs from the first and second observing runs of Advanced LIGO and Advanced Virgo. We find a negative correlation between mass and the mean effective spin (χ¯eff), and positive correlation with its dispersion (σχeff) at 75\% and 80\% confidence. This trend is robust against the choice of mass variable, but most pronounced when the mass variable is taken to be the chirp mass of the binary. The result is consistent with significant contributions from both dynamically assembled and field binaries in the GWTC-1 catalog. The upcoming LIGO O3a data release will critically test this interpretation.

Machine learning algorithms relying on deep neural networks recently allowed a great leap forward in artificial intelligence. Despite the popularity of their applications, the efficiency of these algorithms remains largely unexplained from a theoretical point of view. The mathematical description of learning problems involves very large collections of interacting random variables, difficult to handle analytically as well as numerically. This complexity is precisely the object of study of statistical physics. Its mission, originally pointed toward natural systems, is to understand how macroscopic behaviors arise from microscopic laws. Mean-field methods are one type of approximation strategy developed in this view. We review a selection of classical mean-field methods and recent progress relevant for inference in neural networks. In particular, we remind the principles of derivations of high-temperature expansions, the replica method and message passing algorithms, highlighting their equivalences and complementarities. We also provide references for past and current directions of research on neural networks relying on mean-field methods.

Ta
2
NiSe
5
is one of the most promising materials for hosting an excitonic insulator ground state. While a number of experimental observations have been interpreted in this way, the precise nature of the symmetry breaking occurring in
Ta
2
NiSe
5
, the electronic order parameter, and a realistic microscopic description of the transition mechanism are, however, missing. By a symmetry analysis based on first-principles calculations, we uncover the discrete lattice symmetries which are broken at the transition. We identify a purely electronic order parameter of excitonic nature that breaks these discrete crystal symmetries and contributes to the experimentally observed lattice distortion from an orthorombic to a monoclinic phase. Our results provide a theoretical framework to understand and analyze the excitonic transition in Ta_2NiSe_5 and settle the fundamental questions about symmetry breaking governing the spontaneous formation of excitonic insulating phases in solid-state materials.

We show how a large family of interacting nonequilibrium phases of matter can arise from the presence of multiple time-translation symmetries, which occur by quasiperiodically driving an isolated, quantum many-body system with two or more incommensurate frequencies. These phases are fundamentally different from those realizable in time-independent or periodically driven (Floquet) settings. Focusing on high-frequency drives with smooth time dependence, we rigorously establish general conditions for which these phases are stable in a parametrically long-lived “preheating” regime. We develop a formalism to analyze the effect of the multiple time-translation symmetries on the dynamics of the system, which we use to classify and construct explicit examples of the emergent phases. In particular, we discuss time quasicrystals which spontaneously break the time-translation symmetries, as well as time-translation symmetry-protected topological phases.

A notion of quantum natural evolution strategies is introduced, which provides a geometric synthesis of a number of known quantum/classical algorithms for performing classical black-box optimization. Recent work of Gomes et al. [2019] on combinatorial optimization using neural quantum states is pedagogically reviewed in this context, emphasizing the connection with natural evolution strategies. The algorithmic framework is illustrated for approximate combinatorial optimization problems, and a systematic strategy is found for improving the approximation ratios. In particular it is found that natural evolution strategies can achieve state-of-art approximation ratios for Max-Cut, at the expense of increased computation time.