>My main criticism is that TrueNorth implements networks of integrate-and-fire spiking neurons...Spiking neurons have binary outputs (like neurons in the brain).
Isn't this better. LeCun is looking at this only from a machine learning perspective.
Machine learning is the goal. This chip should be judged on its learning performance, not on how well it adheres to an oversimplified and incomplete model of how biological neurons might work in the brain.
FWIW, there is no learning on-chip. Machine learning is not the goal of this project, nor is its success dependent on it's learning capabilities (at least not at this phase). Where it does succeed, however, is in low-power computation in an architecture that is scalable and fault tolerant. LeCunn is criticizing an orange for not tasting like an apple.
Learning may not happen on-chip, but the network is still learned, and the performance of the chip is dependent on the learning. The spiking architecture of the chip means that the best learning algorithms can't be used. An ASIC implementing a convolutional neural net could also be low-power, scalable, and fault-tolerant, while taking advantage of the best currently known learning algorithms and ultimately performing a lot better on real tasks.
Just because the brain does it one way, doesn't mean it's the best way. Birds versus jets, nature versus artificial.
That being said, the brain is basically the only working example of true intelligence we have, so perhaps trying to emulate it isn't a bad idea.
But I have this sci-fi notion that eventually the AI community will produce some sort of intelligence that is unimaginably different from our current notion of a brain.
The neurons in the IBM chip are only slightly better approximations to biological behavior, nowhere near close enough to hope that it will accidentally result in brain-like intelligence.
Last I heard, the behavior of biological neurons was so badly understood that even the behavior of the 300-neuron C. elegans worm could not be accurately simulated, even though the neuron connections have been fully determined. That was about 8 years ago, though.
I should note that Blue Brain Project is very critical against the group which produced TrueNorth. The director of Blue Brain Project wrote an open letter titled "IBM's claim is a hoax", and considers the approach completely useless for neuroscience research.
Isn't this better. LeCun is looking at this only from a machine learning perspective.