How do we flexibly categorize objects under changing task requirements? Our new paper in Nature Communications (@serences.bsky.social & @nuttidanuttida.bsky.social) examines this: www.nature.com/articles/s41... see 🧵👇

But a tricky aspect of real-world tasks is that the rules can change over time (think: categorizing fruits and vegetables differently when you’re making a salad vs. baking a pie). How does the brain flexibly represent objects during dynamic categorization?

To test this, we trained participants to categorize novel shape stimuli along different category boundaries, while in the fMRI scanner. We then used multivariate decoding from retinotopic visual areas to measure shape discriminability in each task.

Our results show representations became more discriminable across task-relevant category boundaries, especially in early visual areas (V1 & V2). These changes were especially noticeable for shapes near the category boundary -- suggesting stronger task effects for perceptually difficult stimuli.

We also found that neural shape discriminability was higher on behaviorally correct vs. incorrect trials, suggesting that early visual representations may directly support task performance.

How does this relate to existing knowledge on feature-based attention? Importantly, our shape task is complex enough that it can't be solved by just paying attention to just one orientation or curvature; it's necessary to integrate information across multiple features & local areas.

In this sense, we see our task as moving closer to naturalistic object categorization, where categories are defined by complex, combinatoric features. This is a key step forward in understanding how top-down signals related to selective attention might operate in real-world visual tasks.