Source author record

Mikhail Kiselev

Mikhail Kiselev appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

cond-mat.stat-mech cond-mat.str-el Neural and Evolutionary Computing cond-mat.mes-hall cond-mat.other Machine Learning

Catalog footprint

What is connected

5works

6topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2022arXiv

Neural Network Optimization for Reinforcement Learning Tasks Using Sparse Computations

This article proposes a sparse computation-based method for optimizing neural networks for reinforcement learning (RL) tasks. This method combines two ideas: neural network pruning and taking into account input data correlations; it makes it possible to update neuron states only when changes in them exceed a certain threshold. It significantly reduces the number of multiplications when running neural networks. We tested different RL tasks and achieved 20-150x reduction in the number of multiplications. There were no substantial performance losses; sometimes the performance even improved.

preprint2022arXiv

Neuromorphic Artificial Intelligence Systems

Modern AI systems, based on von Neumann architecture and classical neural networks, have a number of fundamental limitations in comparison with the brain. This article discusses such limitations and the ways they can be mitigated. Next, it presents an overview of currently available neuromorphic AI projects in which these limitations are overcame by bringing some brain features into the functioning and organization of computing systems (TrueNorth, Loihi, Tianjic, SpiNNaker, BrainScaleS, NeuronFlow, DYNAP, Akida). Also, the article presents the principle of classifying neuromorphic AI systems by the brain features they use (neural networks, parallelism and asynchrony, impulse nature of information transfer, local learning, sparsity, analog and in-memory computing). In addition to new architectural approaches used in neuromorphic devices based on existing silicon microelectronics technologies, the article also discusses the prospects of using new memristor element base. Examples of recent advances in the use of memristors in euromorphic applications are also given.

preprint2014arXiv

Critical Exponents of the Superfluid-Bose Glass Transition in Three-Dimensions

Recent experimental and numerical studies of the critical-temperature exponent $ϕ$ for the superfluid-Bose glass universality in three-dimensional systems report strong violations of the key quantum critical relation, $ϕ=νz$, where $z$ and $ν$ are the dynamic and correlation length exponents, respectively, and question the fundamental concepts underlying quantum critical phenomena. Using Monte Carlo simulations of the disordered Bose-Hubbard model, we demonstrate that previous work on the superfluid-to-normal fluid transition-temperature dependence on chemical potential (or magnetic field, in spin systems), $T_c \propto (μ-μ_c)^ϕ$, was misinterpreting transient behavior on approach to the fluctuation region with the genuine critical law. When the model parameters are modified to have a broad quantum critical region, simulations of both quantum and classical models reveal that the $ϕ=νz$ law [with $ϕ=2.7(2)$, $z=3$, and $ν= 0.88(5)$] holds true, resolving the $ϕ$-exponent "crisis".

preprint2014arXiv

Equilibrium Fermi-liquid coefficients for the fully screened N-channel Kondo model

We analytically and numerically compute three equilibrium Fermi-liquid coefficients of the fully screened $N$-channel Kondo model, namely $c_B$, $c_T$ and $c_\varepsilon$, characterizing the magnetic field and temperature dependence of the resisitivity, and the curvature of the equilibrium Kondo resonance, respectively. We present a compact, unified derivation of the $N$-dependence of these coefficients, combining elements from various previous treatments of this model. We numerically compute these coefficients using the numerical renormalization group, with non-Abelian symmetries implemented explicitly, finding agreement with Fermi-liquid predictions on the order of 5% or better.

preprint2011arXiv

Phase Diagram of Commensurate Two-Dimensional Disordered Bose Hubbard Model

We establish the full groundstate phase diagram of disordered Bose-Hubbard model in two-dimensions at unity filling factor via quantum Monte Carlo simulations. Similarly to the three-dimensional case we observe extended superfluid regions persisting up to extremely large values of disorder and interaction strength which, however, have small superfluid fractions and thus low transition temperatures. In the vicinity of the superfluid--insulator transition of the pure system, we observe an unexpectedly weak---almost not resolvable---sensitivity of the critical interaction to the strength of (weak) disorder.