CHANNELS

Aerospace

Automotive

Automated & Autonomous Vehicles

Battery & Electrification Technology

Defense

Drone TV

Electronics

Manufacturing & Prototyping

Materials

Medical

Motion Control

RF & Microwave Technology

Robotics & Automation

Sensors & IoT

Test & Measurement

Aerospace
Automotive
Automated & Autonomous Vehicles
Battery & Electrification Technology
Defense
Drone TV
Electronics
Manufacturing & Prototyping
Materials
Medical
Motion Control
RF & Microwave Technology
Robotics & Automation
Sensors & IoT
Test & Measurement
Aerospace & Defense
Search Results

Mini-Labs to Test AI Algorithms

ETH mathematician Juan Gamella has built special “mini-labs” that can be used as test beds for new AI algorithms. Watch this video to see Gamella present his new approach that researchers can use to check how reliably their algorithms and AI models work.

“The mini-labs provide a flexible test environment that delivers real measurement data," says ETH mathematician, Juan Gamella  . "They’re a bit like a playground for algorithms, where researchers can test their AI beyond simulated data in a controlled and safe environment.”

Topics:
AR/AI Test & Measurement

More From SAE Media Group

    Transcript

    00:00:01 a lot of the research that we do in AI revolves around creating intelligent agents that can interact with their environment for example to help human scientists make new discoveries or to efficiently control complex industrial processes my name is hanga and I'm a last year PhD student at eth Zurich where I do research in artificial intelligence the way researchers in AI

    00:00:26 test their work is using computer simulations of reality and this has many many advantages but the problem is that sometimes these simulations fail to imitate some subtle but important aspects of reality and this can give us a false perception of how well these agents actually perform now of course before we deploy these agents in the real world we need

    00:00:48 to make sure that they work for example uh you wouldn't trust uh a pilot was only flown in a computer simulation no matter how good the simulation is and the problem is that the real world is potentially a very expensive and dangerous place to try things out so to give researchers a stepping stone between computer simulations and complex real world scenarios I've have built

    00:01:11 these two devices which I call the causal Chambers and they are miniature computer control Laboratories built around very well understood physical mechanisms what this allows us to do is to test our algorithms in a real but controlled environment providing a sanity check for the work that we do in AI so this is one of the two Chambers which is a miniature wind tunnel so we

    00:01:33 have two fans that push air through it and an additional opening that connects it to the outside a challenge that we might give an AI agent is that by itself it should learn the Dynamics of how the wind tunnel works and use them for example to control the air pressure inside the tunnel of course this is only an example and how the chambers are actually used will depend on the kind of

    00:01:54 algorithms and agents that we're trying to test the response I've had so far from the research community has been fantastic and the chambers are already in use by researchers around the world I'm developing new Chambers built around different physics and I'm working to make them available to as many researchers as possible I truly believe

    00:02:16 that they stand to substantially improve the quality of research that we do in AI closing the gap between research and application