Multi-Agent Deep Reinforcement Learning-Based Resource Allocation in HPC/AI Converged Cluster

As the complexity of deep learning(DL)networks and training data grows enormously,methods that scale with computation are becoming the future of artificial intelligence(AI)development.In this regard,the interplay between machine learning(ML)and high-performance computing(HPC)is an innovative paradigm to speed up the efficiency of AI research and development.However,building and operating an HPC/AI converged system require broad knowledge to leverage the latest computing,networking,and storage technologies.Moreover,an HPC-based AI computing environment needs an appropriate resource allocation and monitoring strategy to efficiently utilize the system resources.In this regard,we introduce a technique for building and operating a high-performance AI-computing environment with the latest technologies.Specifically,an HPC/AI converged system is configured inside Gwangju Institute of Science and Technology(GIST),called GIST AI-X computing cluster,which is built by leveraging the latest Nvidia DGX servers,high-performance storage and networking devices,and various open source tools.Therefore,it can be a good reference for building a small or middlesized HPC/AI converged system for research and educational institutes.In addition,we propose a resource allocation method for DL jobs to efficiently utilize the computing resources with multi-agent deep reinforcement learning(mDRL).Through extensive simulations and experiments,we validate that the proposed mDRL algorithm can help the HPC/AI converged cluster to achieve both system utilization and power consumption improvement.By deploying the proposed resource allocation method to the system,total job completion time is reduced by around 20%and inefficient power consumption is reduced by around 40%.

Identification of Neuro-Fuzzy Hammerstein Model Based on Probability Density Function

A new identification method of neuro-uzzy Hammerstein model based on probability density function(PDF) is presented,which is different from the idea that mean squared error(MSE) is employed as the index function in traditional identification methods.Firstly,a neuro-fuzzy based Hammerstein model is constructed to describe the nonlinearity of Hammerstein process without any prior process knowledge.Secondly,a kind of special test signal is used to separate the link parts of the Hammerstein model.More specifically,the conception of PDF is introduced to solve the identification problem of the neuro-fuzzy Hammerstein model.The antecedent parameters are estimated by a clustering algorithm,while the consequent parameters of the model are identified by designing a virtual PDF control system in which the PDF of the modeling error is estimated and controlled to converge to the target.The proposed method not only guarantees the accuracy of the model but also dominates the spatial distribution of PDF of the model error to improve the generalization ability of the model.Simulated results show the effectiveness of the proposed method.

Joint Multi-modal Parcellation of the Human Striatum:Functions and Clinical Relevance

The human striatum is essential for both lowand high-level functions and has been implicated in the pathophysiology of various prevalent disorders,including Parkinson's disease(PD)and schizophrenia(SCZ).It is known to consist of structurally and functionally divergent subdivisions.However,previous parcellations are based on a single neuroimaging modality,leaving the extent of the multi-modal organization of the striatum unknown.Here,we investigated the organization of the striatum across three modalities—resting-state functional connectivity,probabilistic diffusion tractography,and structural covariance—to provide a holistic convergent view of its structure and function.We found convergent clusters in the dorsal,dorsolateral,rostral,ventral,and caudal striatum.Functional characterization revealed the anterior striatum to be mainly associated with cognitive and emotional functions,while the caudal striatum was related to action execution.Interestingly,significant structural atrophy in the rostral and ventral striatum was common to both PD and SCZ,but atrophy in the dorsolateral striatum was specifically attributable to PD.Our study revealed a cross-modal convergent organization of the striatum,representing a fundamental topographical model that can be useful for investigating structural and functional variability in aging and in clinical conditions.