iiinova — where your digital self keeps evolving

Cheng Xue @cxue.bsky.social · 15/09/2026

This was a team effort! Sol Markman, Ruoyi Chen, and Lily Kramer were undergrads and postbacs when working on this project. They are now future neuro PhDs at MIT, Caltech, and UChicago! Unconditional support from @marlenecohen.bsky.social made this work possible. See her post at tinyurl.com/yjmuj5m9

tinyurl.com

Marlene Cohen (@marlenecohen.bsky.social)

Now out in Nature Neuroscience: fantastic work by @cxue.bsky.social, Sol Markman et al who discovered a neural reason that performance suffers under cognitive load: “Feature interference underlies a n...

050

Cheng Xue @cxue.bsky.social · 15/09/2026

Our inability to multitask isn’t lack of willpower or discipline: our brain is prone to interference from irrelevant info when switching. Accepting that leaves us with a simple solution: tackle one task until it's finished, and give the brain a brief buffer time to turn to another.

110

Cheng Xue @cxue.bsky.social · 15/09/2026

Just several of the possible next steps: the neuronal mechanism behind different cognitive flexibility decline in different disorders may help diagnosis and intervention; attention's role as an active shield against representational entanglement; and reliable decoders for BCI.

100

Cheng Xue @cxue.bsky.social · 15/09/2026

We validated this hypothesis with behavioral, electrophysiological, and causal data. These lines of evidence all point to feature interference as a neuronal limit to flexible task switching.

100

Cheng Xue @cxue.bsky.social · 15/09/2026

Irrelevant information lingers on under task certainty, which interferes more with the relevant information, biasing task performance in a specific direction. We call it feature interference.

120

Cheng Xue @cxue.bsky.social · 15/09/2026

In one model, when task is certain, hidden state evolves into one of two attractors, each represents a choice in that task, separated by a saddle. Task-irrelevant info quickly decays on the descent in this case, but not when the task is uncertain.

100

Cheng Xue @cxue.bsky.social · 15/09/2026

So we trained an RNN to replicate the monkeys' choices, errors included. This network then served as our specimen to generate hypothesis about the actual neuronal mechanism.

100

Cheng Xue @cxue.bsky.social · 15/09/2026

Task switching comes with a drop in the performance of the chosen task, as well as slower responses and lower confidence. We hypothesized that the timing and type of error might reveal the brain's internal mechanism.

100

Cheng Xue @cxue.bsky.social · 15/09/2026

We tracked the participants judgment and confidence regarding the task they chose to do, and how well / confident they completed the chosen task.

110

Cheng Xue @cxue.bsky.social · 15/09/2026

Cognitive flexibility declines with aging and various disorders. Even the healthy and young often feel stretched to the limit between multiple tasks. Why so hard? We investigated how monkeys, humans, and RNNs switch between location and frequency discrimination tasks, with implicitly changing rules.

100

Cheng Xue @cxue.bsky.social · 15/09/2026

Our work recently online in Nature Neuroscience: a neural reason why we err more when juggling tasks nature.com/articles/s41593-026-02430-w 🧵 Switching between tasks for different goals is an important form of cognitive flexibility.

13517

Cheng Xue @cxue.bsky.social · 01/06/2026

Apologies for the display issue with some pictures in the post. Corrected ones here below.

010

Cheng Xue @cxue.bsky.social · 01/06/2026

...which makes this a good paper to cite when you tell your parents that you don't need to clean your room. bsky.app/profile/gouk... Acknowledgements to @simonsfoundation.org, @marlenecohen.bsky.social for supporting my work. #Neuroscience #CognitiveFlexibility #NeuralGeometry #ScienceSky

440

Cheng Xue @cxue.bsky.social · 01/06/2026

Future findings on task-dependent computation may bring about important changes in basic sciences and in clinical translations. Perhaps the brain was never designed to be neat; it is evolved to be flexible enough to survive the complex world, even that means being a little messy,

120

Cheng Xue @cxue.bsky.social · 01/06/2026

Tolstoy famously wrote: "Each unhappy family is unhappy in its own way". The same seems true for malfunctions of the brain's task management system under aging and different neuropsychiatric conditions. Data-driven approaches may be powerful tools to identify their unique mechanistic origins.

110

Cheng Xue @cxue.bsky.social · 01/06/2026

2. Embrace the mess An emerging body of data-driven models are built to capture real behavioral data or neural activity. For instance, error patterns may reveal the underlying neural geometry/computation that generated them. Models that learned it offer a gateway to the underlying mechanism.

110

Cheng Xue @cxue.bsky.social · 01/06/2026

All biological systems operate under constraints and compromises: metabolically, anatomically, ecologically... Finding specific factors that explains the particularities of task management in the behavioral and neuronal level is crucial to understand the computational principles.

110

Cheng Xue @cxue.bsky.social · 01/06/2026

We discussed two approaches forward. 1. Expand the "neat". The sculpture "Schwerpunkt" at MIT McGovern demonstrates how a seemingly messy scatter of neurons neatly form a brain shape when viewed from the correct angle. A better "neat" theory may resolve the apparent "mess".

110

Cheng Xue @cxue.bsky.social · 01/06/2026

But in reality, things can look disturbingly "messy". Task learning / switching is slow and imperfect. Complex task implementations are not as modularized as they can be. Task relevant and irrelevant information often gets entangled. Are these features or bugs of the brain?

150

Cheng Xue @cxue.bsky.social · 01/06/2026

This "neat" feature is important because disentangled neural task spaces can serve as building blocks for complex functions, which can be recombined to help the organism adapt to new situations and even solve problems it has never encountered before.

130

Cheng Xue @cxue.bsky.social · 01/06/2026

So it is tempting to believe the brain is handling multiple tasks neatly. Different tasks occupy distinct neural manifolds without interference, which allow clean transitions when needed: just like how an operating system allots distinct memory space for multiple programs.

120

Cheng Xue @cxue.bsky.social · 01/06/2026

We handle multiple tasks effortlessly, often without realizing it. Consider driving. You decide to hit gas or brake; and the next moment you decide to turn left or right. This smooth juggling of different computations in the brain is an important form of cognitive flexibility.

130

Cheng Xue @cxue.bsky.social · 01/06/2026

Excited to share our new review in Trends in Neuroscience with Gouki Okazawa @gouki-okazawa.bsky.social : "The 'neat' and 'messy' in task-dependent neural geometry and computation" where we reviewed what we know about the 'task-manager' in the brain. A thread 👇

13212