Los Alamos National LaboratoryInformation Science and Technology Institute (ISTI)
Implementing and fostering collaborative research, workforce and program development, and technical exchange

Seminar Series

The IS&T seminars on various information science and technology topics are held every Wednesday from 3-4 PM at the CNLS Conference Room unless otherwise noted.

Contact  

  • Institute Director
  • Stephan Eidenbenz
  • (505) 667-3742
  • Email
  • Institute Deputy Director
  • Kary Myers
  • (505) 606-1455
  • Email
  • Professional Staff Assistant
  • Nickole Aguilar Garcia
  • (505) 665-3048
  • Email

To schedule a speaker please contact Nickole Garcia.

To subscribe to the IS&T seminar announcements please use the Program Announcement tool (use win domain password to login)/ Select the "NSEC - Information Science and Technology Institute" to receive announcements.

March 28, 2018: Patrick Taylor, Missouri University of Science and Technology

Effective Learning and Control in Biologically Realistic Spiking Neural Network Simulations

Abstract

Many varieties of neural networks excel in AI and machine learning applications. Some types of problems remain difficult for neural networks to solve, particularly rapidly time-varying stimuli such as speech or sensory processing for control. Learning rules are well-developed for feed-forward, deep, convolutional, and simple recurrent networks, which perform well for relatively static or step-wise problems such as image recognition or the game of Go. Yet, despite much desire to relate deep convolutional networks to brain function, they are almost entirely biologically unrealistic and unrepresentative of brain activity and processing. On the other hand, learning rules for naturally recurrent spiking neural networks have proven exceptionally difficult, but these models are much more representative of biological function in brains. Though few supervised learning rules for spiking networks have shown success on tasks such as clustering or dimensionality reduction, supervised and reinforcement learning rules remain grossly underdeveloped and unsuccessful. Interestingly, if input rates vary rapidly (around the time scale of the learning window) spike-timing methods can, at least in theory, be distinctly more computationally powerful than rate or sigmoidal neurons. We take a hybrid approach by synthesizing unsupervised learning rules relying only on cell-local information combined with global feedback, to produce a form of general reinforcement learning with embedded clustering. Combining a global reinforcement signal with spike timing dependent plasticity better mimics biological processes, and may out-perform existing learning systems in some types of temporally rich processing or control tasks, or in ability to generalize over problem domains. Rather than using explicit state or action learning, a basic spike timing dependent plasticity (STDP) rule, which performs a kind of asymmetric Hebbian fire-together wire-together, can be modified to perform similarly to TD learning, implicitly via an associative rule with feedback. Notably, these neurons use only cell-local information with global uniform feedback. They are some of the most biologically realistic models of cell-level learning, and mimic biological data on long-term potentiation (LTP) and depression (LTD) at synapses, bulk neurotransmission via dopamine, as well as the ionic eligibility traces necessary to implement such reinforcement learning.

Host: Chris Rawlings, 505-667-4049, crawlings@lanl.gov or Daniel Tauritz, tauritzd@mst.edu

January 23, 2018: Philippe Pebay, NexGen Analytics

Efficient Visualization and Analysis of Large-Scale, Tree-Based, Adaptive Mesh Refinement Simulations with Rectilinear Geometry

Abstract

The goal of this presentation is to explain our adaptive approach for the visualization and analysis of large-scale, parallel, tree-based adaptive mesh refinement (AMR) scientific simulations, using optimized data structures and customized traversal objects, taking advantage of the intrinsic structure of tree-based AMR grids. When compared to those obtained with standard approaches for the visualization of such datasets, our results indicate a gain of at least 80% in terms of memory footprint with better-quality rendering, while retaining similar execution speed. Furthermore, our innovative approach allows for further acceleration of the rendering of 2-dimensional AMR grids, hereby solving the problem posed by the loss of interactivity that occurs when dealing with large and/or deeply refined meshes. This presentation will be illustrated with several examples and performance assessment results.

Host: John Patchett, 505-665-1110, patchett@lanl.gov

Visit Blogger Join Us on Facebook Follow Us on Twitter See our Flickr Photos Watch Our YouTube Videos Find Us on LinkedIn Find Us on iTunesFind Us on GooglePlayFind Us on Instagram