Both chats this week packed a mighty punch of inspiration!
Alison Lowndes joined Datanauts to talk about her work with NVIDIA on deep learning and artificial intelligence.
Many know NVIDIA from its prominent role in the gaming industry as a designer and manufacturer of graphics processing unit (GPU) products. In recent years, NVIDIA’s focus has shifted and expanded to include high performance computing, visualization of large data sets, and artificial intelligence. AI and deep learning research is rapidly expanding.
“There is a staggering amount of funding being driven by, pardon the pun, by the self-driving car industry,” said Alison. “But it’s also essential because the human genome was not built to understand [large datasets].”
Five years ago, some pioneering researchers applied the use of GPUs rather than central processing units (CPUs)—a difference of 5,000 compared to four processing cores, respectively—to analyze big datasets. This landmark paper brought awareness to the practice and availability of using GPUs.
“We are nowhere near human intelligence, "said Alison. "We are probably the equivalent of a three-year-old child at the moment, so we’re a long way off from the crazy killer robots...but we’re certainly superhuman at certain narrow AI tasks.”
Certain benchmarks, such as ImageNet, combine GPUs with a deep neural network architecture running a supervised learning task, the computer could correctly analyze and classify millions of images in a dataset, achieving what Alison referred to as a “Ph.D. level” of intelligence in something like 15 minutes.
Running training datasets through the untrained neural network model in a deep learning framework ultimately produces a trained model that can be deployed wherever, be it a data center, a mobile device, or a self-driving car.
For more dynamic problems, like speech, where training a model requires backward and forward steps, recurrent neural network architectures like the Long Short-Term Memory (LSTM) provide memory “cells” that can remember values over certain time intervals. Apple’s “Quicktype” iPhone function and Siri and Amazon’s Alexa all use LSTM for natural language text compression.
AI is also being used to augment our human bodies by trying to recreate biological and neurological processes using neural networks by transforming “inputs” like neuron pulses into mathematical problems. Applications for this work range from Google’s DeepMind AlphaGo and Parkour, theIssac Lab robot simulator, and even self-driving cars!
Jennifer Lopez, one of the founding Datanauts, shared some of the projects she is working on with the Center for the Advancement of Science inSpace (CASIS), where she is the commercial innovation lead.
In this role, Jennifer helps CASIS and its partners, which include NASA, optimize new technology and operations to maximize utilization of the International Space Station for terrestrial benefit with a goal toward commercialization of low Earth orbit and success of future orbiting platforms.
Many of the projects developed by CASIS take advantage of the International Space Station’s unique position in a microgravity environment with temperature extremes and with an advantageous vantage point of Earth. The projects below have recently launched or will launch soon to the ISS:
· Anna-Sophia Boguraev: Won the inaugural Genes in Space competition in 2015 and sent her own DNA experiment to the ISS to evaluate how any changes to astronauts’ genetic sequences might be related to immune system responses to the microgravity environment of space.
Jennifer and CASIS also work closely with NASA on a few other projects, including SPHERES and its successor Astrobee. Both are free-flying, bowling ball-sized vehicles that operate inside the ISS to assist astronauts with “household” tasks and provide observations of the ISS environment for NASA engineers.
Jennifer will have some exciting updates for the next class of Datanauts, so keep an eye out for those!
Applications for the Spring 2018 are open now!
Ronnie has been enthusiastically showcasing NASA data as a member of NASA's Open Data team since 2013. She supports NASA's open source efforts by helping to curate and administrate datasets on NASA's Open Data Portal and Open Source Code Catalog, managing citizen and internal requests for NASA data, contributing to the Space Data Daily Open NASA blog, teaching Datanauts courses, and coordinating logistics and data support for the International Space Apps Challenge hackathons.