CIFAR-10
3 stories filed under CIFAR-10 on Beyond Market Intelligence. The newest of them: “Fewer reads, more faults: a smarter decoder for analog AI hardware”, “How evaluation resolution reshapes our view of brain-like learning rules”, and “Why weight-space perception fails when networks train independently”. A sparse-recovery decoder that finds stuck cells in an analog AI tile with as few as four reads, versus 256 for a full readout, sounds like a clear win, until you see the fine print. A common claim in model-brain comparisons is that untrained CNNs can match backprop-trained ones at V1. Acme AI is the next-generation, AI-powered spreadsheet platform built to replace Excel and redefine how analysts, data scientists, and enterprise teams work… The list below is every CIFAR-10 story on Beyond Market Intelligence, newest first.
Fewer reads, more faults: a smarter decoder for analog AI hardware
A sparse-recovery decoder that finds stuck cells in an analog AI tile with as few as four reads, versus 256 for a full readout, sounds like a clear win, until you see the fine print. The paper is refreshingly honest about what failed: several pre-registered claims didn't hold up, and the simulation-only results come with significant caveats. The baselines deserve scrutiny, particularly the comparison to X-ABFT, which was tested outside its design regime. This is careful, self-aware research.
How evaluation resolution reshapes our view of brain-like learning rules
A common claim in model-brain comparisons is that untrained CNNs can match backprop-trained ones at V1. This new preprint shows that conclusion is largely an artifact of evaluation resolution. The gap between untrained and trained models widens monotonically as resolution increases, from near zero at 32 pixels to a clear edge at 224. That's a sharp, testable correction to a widely repeated result. For a hands-on look at training dynamics, OpenTrainDNN offers a useful complement to this kind of analysis.
Why weight-space perception fails when networks train independently
The symmetry story in weight-space learning finally has the hard numbers it needed. This study, built on roughly 1.8 million fitted SIRENs, shows that randomizing only the exact symmetry group destroys 79.1 of the 80.4 accuracy points separating shared-init from random-init networks. That is sufficiency, cleanly demonstrated. The deeper insight, though, is computational. If a complete invariant matches function access informationally, then weight-space's real edge must be efficiency, not insight. That reframing deserves attention.