It may not be. I’ve been out of the field for a long time. But my understanding is so far they’ve done white box, noise, and black box transfer—-I’d like to see black box direct optimization, so SPSA or a newer better alternative
@goodfellow_ian
-
Bhattacharya Great Barrington Declaration Public Health Board
By
–
As a key signatory of the Great Barrington Declaration, Bhattacharya has been instrumental in dismantling public health worldwide. It would be hard to think of a worse board member for your stated mission.
-
Adversarial Training Benchmarking: Average Case vs Attacker Strategy
By
–
As one of the co-inventors of adversarial training I endorse Carlini’s take. The issue is not so much which attack transformations are allowed, the issue is using average case benchmarks when an attacker will not randomly sample from average starting points
-
Max Welling’s Carbon Capture Materials Science Startup Launch
By
–
I don’t know for sure if this will be discussed, but I’m excited for Max Welling’s materials science startup, focusing on carbon capture
-
Theano developers CUDA neural networks training infrastructure history
By
–
I wouldn’t be surprised at all if for example one of the Theano developers like James Bergstra had been training neural nets on CUDA before me though. Software / infra was undervalued in academia and just not published nearly as much
-
First Neural Networks Training on NVIDIA CUDA GPU
By
–
I definitely wasn’t the first to do neural nets on GPU and the MSR work is the earliest I know of. I might’ve been the first to train neural nets on *NVIDIA CUDA* if we care about brand history rather than technical approach
-
Low Temperature Sampling Improves AI Decision Making Accuracy
By
–
I haven’t read the paper but I think the basic idea is that the shots that hit the target tend to be in the mode of the distribution and the shots that miss tend to be in the tails of the distribution, so low temperature sampling discards more bad moves than good
-
Low temperature sampling technique in AI model research
By
–
That’s not literally the idea in the paper just an intuition builder. The paper is more about how you can get a similar effect via low temperature sampling
-
Learning from Success: Imitation Learning in AI Systems
By
–
The idea might be easier to understand for something that’s more of a motor skill like archery. Imagine building a dataset of humans shooting arrows at targets and then imitating only the examples where they hit the targets.