Search

Interesting data often concentrate on low dimensional smooth manifolds inside a high dimensional ambient space. Random projections are a simple, powerful tool for dimensionality reduction of such data. Previous works have studied bounds on how many projections are needed to accurately preserve the geometry of these manifolds, given their intrinsic dimensionality, volume and curvature. However, such works employ definitions of volume and curvature that are inherently difficult to compute. Therefore such theory cannot be easily tested against numerical simulations to understand the tightness of the proven bounds. We instead study typical distortions arising in random projections of an ensemble of smooth Gaussian random manifolds. We find explicitly computable, approximate theoretical bounds on the number of projections required to accurately preserve the geometry of these manifolds. Our bounds, while approximate, can only be violated with a probability that is exponentially small in the ambient dimension, and therefore they hold with high probability in cases of practical interest. Moreover, unlike previous work, we test our theoretical bounds against numerical experiments on the actual geometric distortions that typically occur for random projections of random smooth manifolds. We find our bounds are tighter than previous results by several orders of magnitude., Comment: 45 pages, 9 figures

We combine Riemannian geometry with the mean field theory of high dimensional chaos to study the nature of signal propagation in generic, deep neural networks with random weights. Our results reveal an order-to-chaos expressivity phase transition, with networks in the chaotic phase computing nonlinear functions whose global curvature grows exponentially with depth but not width. We prove this generic class of deep random functions cannot be efficiently computed by any shallow network, going beyond prior work restricted to the analysis of single functions. Moreover, we formalize and quantitatively demonstrate the long conjectured idea that deep networks can disentangle highly curved manifolds in input space into flat manifolds in hidden space. Our theoretical analysis of the expressive power of deep networks broadly applies to arbitrary nonlinearities, and provides a quantitative underpinning for previously abstract notions about the geometry of deep functions., Comment: Fixed equation references

Maximizing the speed and precision of communication while minimizing power dissipation is a fundamental engineering design goal. Also, biological systems achieve remarkable speed, precision and power efficiency using poorly understood physical design principles. Powerful theories like information theory and thermodynamics do not provide general limits on power, precision and speed. Here we go beyond these classical theories to prove that the product of precision and speed is universally bounded by power dissipation in any physical communication channel whose dynamics is faster than that of the signal. Moreover, our derivation involves a novel connection between friction and information geometry. These results may yield insight into both the engineering design of communication devices and the structure and function of biological signaling systems., Comment: 15 pages, 3 figures

Recent experimental advances in neuroscience have opened new vistas into the immense complexity of neuronal networks. This proliferation of data challenges us on two parallel fronts. First, how can we form adequate theoretical frameworks for understanding how dynamical network processes cooperate across widely disparate spatiotemporal scales to solve important computational problems? And second, how can we extract meaningful models of neuronal systems from high dimensional datasets? To aid in these challenges, we give a pedagogical review of a collection of ideas and theoretical methods arising at the intersection of statistical physics, computer science and neurobiology. We introduce the interrelated replica and cavity methods, which originated in statistical physics as powerful ways to quantitatively analyze large highly heterogeneous systems of many interacting degrees of freedom. We also introduce the closely related notion of message passing in graphical models, which originated in computer science as a distributed algorithm capable of solving large inference and optimization problems involving many coupled variables. We then show how both the statistical physics and computer science perspectives can be applied in a wide diversity of contexts to problems arising in theoretical neuroscience and data analysis. Along the way we discuss spin glasses, learning theory, illusions of structure in noise, random matrices, dimensionality reduction, and compressed sensing, all within the unified formalism of the replica method. Moreover, we review recent conceptual connections between message passing in graphical models, and neural computation and learning. Overall, these ideas illustrate how statistical physics and computer science might provide a lens through which we can uncover emergent computational functions buried deep within the dynamical complexities of neuronal networks., Comment: 72 pages, 8 figures, iopart.cls, to appear in JSTAT

We use the AdS/CFT correspondence in a regime in which the field theory reduces to fluid dynamics to construct an infinite class of new black objects in Scherk-Schwarz compactified AdS(d+2) space. Our configurations are dual to black objects that generalize black rings and have horizon topology S^(d-n) x T^n, for n less than or equal to (d-1)/2. Locally our fluid configurations are plasma sheets that curve around into tori whose radii are large compared to the thickness of the sheets (the ratio of these radii constitutes a small parameter that permits the perturbative construction of these configurations). These toroidal configurations are stabilized by angular momentum. We study solutions whose dual horizon topologies are S^3 x S^1, S^4 x S^1 and S^3 x T^2 in detail; in particular we investigate the thermodynamic properties of these objects. We also present a formal general construction of the most general stationary configuration of fluids with boundaries that solve the d-dimensional relativistic Navier-Stokes equation., Comment: 32 pages, 10 figures; v2: Some figures are modified and a few references added.

We use the AdS/CFT correspondence to argue that large rotating black holes in global AdS(D) spaces are dual to stationary solutions of the relativistic Navier-Stokes equations on S**(D-2). Reading off the equation of state of this fluid from the thermodynamics of non-rotating black holes, we proceed to construct the nonlinear spinning solutions of fluid mechanics that are dual to rotating black holes. In all known examples, the thermodynamics and the local stress tensor of our solutions are in precise agreement with the thermodynamics and boundary stress tensor of the spinning black holes. Our fluid dynamical description applies to large non-extremal black holes as well as a class of large non-supersymmetric extremal black holes, but is never valid for supersymmetric black holes. Our results yield predictions for the thermodynamics of all large black holes in all theories of gravity on AdS spaces, for example, string theory on AdS(5) x S**5 and M theory on AdS(4) x S**7 and AdS(7) x S**4., Comment: 62 pages, 1 figure. v2: references, typos

We construct solutions to the relativistic Navier-Stokes equations that describe the long wavelength collective dynamics of the deconfined plasma phase of N=4 Yang Mills theory compactified down to d=3 on a Scherk-Schwarz circle and higher dimensional generalisations. Our solutions are stationary, axially symmetric spinning balls and rings of plasma. These solutions, which are dual to (yet to be constructed) rotating black holes and black rings in Scherk-Schwarz compactified AdS(5) and AdS(6), and have properties that are qualitatively similar to those of black holes and black rings in flat five dimensional supergravity., Comment: 40 pages, 40 figures. (v2) Correction to black brane equation of state, additional references

Mikhailov has constructed an infinite family of 1/8 BPS D3-branes in AdS(5) x S**5. We regulate Mikhailov's solution space by focussing on finite dimensional submanifolds. Our submanifolds are topologically complex projective spaces with symplectic form cohomologically equal to 2 pi N times the Fubini-Study Kahler class. Upon quantization and removing the regulator we find the Hilbert Space of N noninteracting Bose particles in a 3d Harmonic oscillator, a result previously conjectured by Beasley. This Hilbert Space is isomorphic to the classical chiral ring of 1/8 BPS states in N=4 Yang-Mills theory. We view our result as evidence that the spectrum of 1/8 BPS states in N=4 Yang Mills theory, which is known to jump discontinuously from zero to infinitesimal coupling, receives no further renormalization at finite values of the `t Hooft coupling., Comment: 66 pages, Latex. v2: typos fixed, references added. v3: more typos and references