Ling-Qi Zhang

@lingqiz.bsky.social

🧠 Comp Neuro, Behavior, Theory and Machine Learning | Theory Fellow @HHMI Janelia | 💻 https://lingqiz.github.io/

It’s very fun to analyze other people’s data for your favorite model and theory, but data reuse actually takes a huge amount of effort. We tested if current agentic AI can help with data reuse. Conclusion: it definitely can, but requires careful human supervision. #AIforScience #compneuro #neuroAI

Kristin Branson @kristinmbranson.bsky.social · 3mo ago

New preprint with @lingqiz.bsky.social: Neurodata Without Boredom: Benchmarking Agentic AI for Data Reuse arxiv.org/abs/2605.12808 1/10

Overview of the data conversion task. The benchmark includes eight datasets spanning a range of neural recording modalities, behavioral tasks, measurements, experiment protocols, and data formats. For each dataset, agents were also given the released paper, methods, and code, together with a structured prompt. The agent’s goal was to convert each heterogeneous source dataset into a common format suitable for a neural decoding task. 
Diagram has a column for each of 8 papers from mouse systems neuroscience. The first row shows the name of the dataset and a description of the focus of that work: 1) Allen2P: Visual change detection,2) Lee2025: Spatial coding & remapping, 3) Majnik2025: Cortical development, 4) Sosa2024: Spatial reward learning,  5) Chen2024: Memory-guided decision, 6) Hasnain2024: Context-based decision, 7) Zhang2025: Visual-motor decision, 8) Zhong2025: Visual learning. 
The next row shows the type of neural recording data: 1, 3, 4, 8: head-fixed calcium imaging, 2: miniprobe, 5-7: neuropixels or other silicon probes
The next row shows the types of behavioral and task variables. These are different for every task
The next row shows a visualization of the experiment protocol for each task. Again, these are different for every task
The next row shows that this information is all condensed into data, paper, and code. Data has the different types of information in different orders and structures. It also shows the type of files: 1: NWB files & SDK, 2,6: MATLAB files, 3, 7: Python files, 4-5: NWB files, 8: Various + API
The next part of the diagram shows the task, which is for agents to convert the data to a common data format for a downstream decoder task.

After three years of working on this review with my collaborator Emily Dennis (Janelia Farm), it is finally in press. The review provides an overview and our personal reflections on what constitutes an integrative approach to neurobiology of foraging. Enjoy www.annualreviews.org/content/jour...

Neurobiology of Foraging: An Integrative Approach

Foraging, defined as the search for food to sustain one's energetic needs, is a fundamental behavior performed by almost all animals to survive in their environment. Foraging involves a variety o...

annualreviews.org

After three years , this joint review with my dear friend Emily Jane Dennis is finally in press. The review provides a glimpse on what constitutes an integrative approach to neurobiology of foraging. Out in the next issue of Annual Reviews of Neuroscience. www.annualreviews.org/content/jour...

Neurobiology of Foraging: An Integrative Approach

Foraging, defined as the search for food to sustain one's energetic needs, is a fundamental behavior performed by almost all animals to survive in their environment. Foraging involves a variety o...

annualreviews.org

These workshops are amazing opportunities for deep discussions on how cognition is implemented in the brain, from theory to behavior and circuits. Apply by May 7 & spread the word!

Janelia Conferences@janeliaconf.bsky.social · 6mo ago

Apply by May 7 for two early-career #neuro workshops at Janelia! Featuring joint sessions to foster collaboration between theory & experiment. 🧠 🤝 Theoretical Neuroscience→ janelia.news/THE26 Mechanistic Cognitive Neuroscience→ janelia.news/CNW26 @ratecoding.bsky.social @jvoigts.bsky.social

The debate of whether the brain is optimal is the wrong one. Quoting one of my favorite writings from Herbert Simon (1992): The predictions of an optimizing theory depend as much on the postulated side conditions as on the optimization assumption. (By side conditions here, he meant constraints).

John Pearson@jmxpearson.bsky.social · 11mo ago

I mean, efficient coding and optimal control work *far* better than they have a right to, and they generate predictions. They are not falsifiable hypotheses except in the narrowest sense, but they are programs that can identify parsimonious principles for explaining real observations.

Columbia University has failed to defend academic freedom & freedom of speech & allowed a great institution to be bullied & extorted into submission by an administration hellbent on contorting higher education to their will. The AAUP condemns this capitulation. We can & must fight back.

Columbia Concedes to Trump’s Demands After Federal Funds Are Stripped

The administration has moved to cut $400 million in federal funding to the university without changes to its policies and rules.

nytimes.com

old Soviet joke for our times: Guy stops by the newsstand every day, scans the front page, doesn’t buy the paper. One day the vendor asks what he’s up to. Guy says: “looking for an obituary.” Vendor says “those are towards the back of the paper, comrade.” Guy says: “not the one I’m looking for.”