a while back i threatened to share this. finally online for detection tasks we often systematically estimates sensitivity wrong. we need to control for unequal variance in models, but we often don't coz it needs extra data now there's a virtually 'free' way to do it www.cell.com/iscience/pdf...

The visual world is composed of objects, and those objects are composed of features. But do VLMs exploit this compositional structure when processing multi-object scenes? In our 🆒🆕 #ICLR2026 paper, we find they do – via emergent symbolic mechanisms for visual binding. 🧵👇

How do you know how visual stimuli are represented internally for decision making? This is perhaps the central question in perceptual decision making. In a new paper, we show that one can use artificial neural networks to crack this problem. #NeuroAi #VisionScience direct.mit.edu/opmi/article...

When reading a line like "the glow of the flame briefly illuminated his face", most readers tend to build an image in their mind. 🧠 People with aphantasia, however, report that they do not. We wanted to know: Does this change how they physically read stories? (1/5)