Why it matters
Prathyush Sajith's talk on compressing stereo-depth transformer models is a concrete example of inference-efficiency research that ultimately lowers per-request compute cost.
The tokenmaxxing angle
The Walsh-Hadamard Transform talk on shrinking stereo pipelines, plus Jayakumar Ramalingam's production-testing talk on drift detection, both map onto tokenmaxxing's interest in cheaper, more reliable inference at scale.
From the organizers
The agenda names Rush University's Mattia Perrone on spine-MRI vision-language models, UIC's Prathyush Sajith on Walsh-Hadamard Transform stereo depth optimization, and SiriusXM's Jayakumar Ramalingam on production testing.