Marlene Cohen

Behavioural and neuronal insights into multisensory combination of unpracticed cues. Effective decision-making requires integrating multiple information sources, weighted by their reliability and context. While classic studies show near-optimal cue combination with well-learned signal...

We’re excited about the connections in this work, between behavior, species, individual differences, brain areas, and neuronal mechanisms. We would love your feedback!
www.biorxiv.org/content/10.1...

Keon is on the job market (ideally in Canada) & has incredibly exciting plans for his future lab. He will use neurophysiology-inspired psychophysics to study how perception and cognition differ across the lifespan and across neurotypical and neurodiverse people. Hire him! www.keonallen.com 8/

This study reflects Keon’s pioneering spirit. He came to our visual neurophysiology lab from a background in psychology and haptic perception and built bridges between fields, from Bayesian models and online behavior to neuronal mechanisms of cue integration. 7/

Together, these results suggest that the brain combines information differently within and across sensory modalities, perhaps from different circuitry in sensory & association areas. These distinctions seem to be conserved across species, and deviations could be diagnostic for brain differences. 6/

Doug conducted parallel neurophysiological experiments in which a visual cue was combined with a causal manipulation.
Electrical microstimulation in visual cortex was integrated with sensory motion cues.
Stimulation in prefrontal cortex instead pushed choices toward winner-take-all. 5/

These strategies varied across people. Age and self-reported ADHD or Autism influenced which cues were judged most accurately and how they were integrated, suggesting that individual differences in multisensory combination may reflect broader cognitive or neural traits. 4/

Keon measured cue combination in large cohorts of neurodiverse participants who made judgments based on multiple cues.
People combined two visual cues nearly optimally. When vision and sound conflicted, behavior became winner-take-all, usually but not always favoring the more reliable cue. 3/

Many studies test how subjects combine information from well-practiced cues with feedback. But we often need to combine unfamiliar signals. For example, we might try to match what we see and hear when a new appliance beeps.
Keon and Doug Ruff asked how brains do that. 2/

Behavioural and neuronal insights into multisensory combination of unpracticed cues. Effective decision-making requires integrating multiple information sources, weighted by their reliability and context. While classic studies show near-optimal cue combination with well-learned signal...

New preprint from the lab! 🎉
Postdoc Keon Allen led this study exploring how subjects combine sources of sensory information, including unfamiliar & cross-modal cues, and what that can tell us about decision-making, neural mechanisms, and neurodiversity.
www.biorxiv.org/content/10.1... 🧵🧪1/

I’m sorry that I won’t be in San Diego to join you. But I hope it’s wonderful, and a heartfelt thank you for everything you do for our community.

Thank you! Your work was definitely the inspiration for a lot of this.

Thanks, Hannah!

Ha! My student was the one who realized we should cite your paper. Maybe she should earn a second PhD in history...

Thank you, especially for the laugh!

Neuronal signatures of successful one-shot memory in mid-level visual cortex High-capacity, one-shot visual recognition memory challenges theories of learning and neural coding because it requires rapid, robust, and durable representations. Most studies have focused on the hip...

We are excited about potential applications of this work, from artificial intelligence to translational efforts to fix memory disorders. This highlights a central value of our field: using curiosity-driven science for broad impact. We’d love your feedback! doi.org/10.1101/2025.09.22.677855 /end

This is the first chapter of Grace’s thesis, and there is so much more to come. She is something special, and I am going to thoroughly enjoy seeing her take our field by storm. 9/

These findings show that the building blocks of fast, high-capacity memory are present in mid-level visual cortex. Take-home: cognition is distributed. And stay tuned: Grace’s next papers will explore mechanisms by which these signals interact with the larger network and are disrupted in disease. 8/

We also found faster response dynamics to familiar images, consistent with pattern completion. This means that after the first couple of image fragments, V4 already signaled the whole image (but only during successful memory). The hippocampus does this, but we were surprised to see it in V4. 🤯 7/

We found all of these neuronal signatures in V4. But the only ones that reliably predicted behavior were related to how consistent population responses were during memory encoding and retrieval. More consistent responses = greater memory success. 6/

We looked for proposed neuronal signatures of memory, including:
• magnitude coding
• repetition suppression
• sparse coding
• population response consistency (=similar responses to novel and familiar images) 5/

Grace and awesome staff scientist Cheng Xue tested whether area V4 contains the signals that could support recognition memory. Their task revealed images bit by bit. This allowed us to analyze dynamics and increased difficulty so we could compare neuronal responses on correct vs error trials. 4/

Most previous studies have focused on hippocampus and higher cortical areas. But behavioral work shows that memorability depends on visual features and recognition memory distinguishes even semantically similar images. Seems like a job for mid-level visual cortex. 3/

When she was a rotation student, Grace DiRisio pointed out that visual recognition memory challenges all our neural coding theories because of its remarkable capacity. Linear codes work for low capacity functions e.g. discrimination & attention. Memory for thousands of images is another story. 2/

Neuronal signatures of successful one-shot memory in mid-level visual cortex High-capacity, one-shot visual recognition memory challenges theories of learning and neural coding because it requires rapid, robust, and durable representations. Most studies have focused on the hip...

New preprint! How can you remember an image you saw once, even after seeing thousands of them? We find a role for humble mid-level visual cortex in high-capacity, one-shot learning. doi.org/10.1101/2025.09.22.677855 🧵🧪1/

We are grateful for sustained federal funding (mostly NIH for us), which is the only thing that makes it possible to work on a problem for decades. This work will translate to people: it suggests targeted treatments for disorders that affect cognition & also correlated variability. Coming soon! /end

First author Ramanujan Srinath demonstrated some of the things that make him a great scientist: he thinks deeply & creatively, brings together many forms of evidence & people, and is determined and innovative. He is on the job market this year and will run an incredible lab – don’t miss out! 5/

Ramanujan Srinath collected & analyzed many data sets including Amy Ni’s, Yunlong (Draco) Xu & Brent Doiron modeled, & Doug did a causal experiment. The key is: when noise and signal come from the same circuit (they must!), then smart, flexible decisions have a strong relationship with noise. 4/

@ramanujan-s.bsky.social and a fantastic team of coauthors figured out a resolution: we propose that correlated variability isn’t noise that corrupts a signal or something to ignore. It reflects the activity in the population that is read out to guide behavior. 3/

a husky puppy is laying on the floor with its tongue out and wearing a blue collar . ALT: a husky puppy is laying on the floor with its tongue out and wearing a blue collar .

Hmm: Correlated variability in sensory neurons (how much responses fluctuate together) is related to behavior: it’s modulated by attention, learning, & motivation and related to individual decisions. But because it's low-D, in theory its effect on behavior should be small. What’s up with that? 2/

Guided by Noise: Correlated Variability Channels Task-Relevant Information in Sensory Neurons Shared trial-to-trial variability across sensory neurons is reliably reduced when perceptual performance improves, yet this variability is low-dimensional, so it could be ignored by an optimal readout...

One of the joys of being a scientist is the ability to think about a problem for a long time. Our new preprint solves a mystery that has been bugging me since I was a graduate student (which was, ahem, a while ago). 🧪🧠🧵1/
www.biorxiv.org/content/10.1...