Neurosalience

“Predictive coding offers a powerful lens for understanding psychosis…”

Dr. Marta Garrido is a professor at the Melbourne School of Psychological Sciences, where she leads the Cognitive Neuroscience and Computational Psychiatry Laboratory and directs the Cognitive Neuroscience Hub. She is also a research program lead at the Graeme Clark Institute. With a background in engineering physics from the University of Lisbon and a PhD in neuroscience from University College London under the mentorship of Professor Karl Friston, Marta has become a leading figure in understanding how the brain processes predictions and surprise. Her research spans mismatch negativity, predictive coding theory, dynamic causal modeling, and the development of cutting-edge neuroimaging technologies, including Australia’s first optically pumped MEG system.

In this episode, Peter and Marta explore the elegant framework of predictive coding and its implications for understanding psychiatric conditions like psychosis. They discuss how the brain generates predictions about sensory input and how disruptions in these mechanisms may contribute to symptoms of mental illness. Marta shares her journey from engineering to neuroscience, her transformative PhD experience, and the challenges of building a new MEG system from the ground up. The conversation covers fascinating topics including mismatch negativity as a prediction error signal, subcortical shortcuts for processing threatening stimuli, the phenomenon of blindsight, and the critical importance of mentorship in academic careers. Marta also offers candid reflections on being a woman in neuroscience and her vision for the future of computational psychiatry.

We hope you enjoy this episode!

Chapters:
00:00 - Introduction to Dr. Marta Guerrero
04:46 - Journey from Engineering to Neuroscience
10:39 - Understanding Predictive Coding and Bayesian Inference
18:34 - Implications of Predictive Coding in Schizophrenia
27:08 - Advancements in Brain Imaging Techniques
36:31 - Exploring Blindsight and Subcortical Shortcuts
44:14 - Reverse Engineering the Brain: Challenges and Ambitions
51:23 - The Journey of Developing Optically Pumped Magnetometers
01:00:29 - Promoting Women in Neuroscience and Leadership Challenges

Works mentioned:
15:59 - Randeniya et al. (2018). Sensory prediction errors in the continuum of psychosis. https://doi.org/10.1016/j.schres.2017.04.019
18:36 - Goodwin et al. (2026). Predictive processing accounts of psychosis: Bottom-up or top-down disruptions. https://doi.org/10.1038/s44220-025-00558-5
26:02 - Larsen et al. (2019). 22q11.2 deletion syndrome: intact prediction but reduced adaptation. https://doi.org/10.1016/j.nicl.2019.101721
29:40 - Garvert et al. (2014). Subcortical amygdala pathways enable rapid face processing. https://doi.org/10.1016/j.neuroimage.2014.07.047
29:40 - McFadyen et al. (2017). A rapid subcortical amygdala route for faces. https://doi.org/10.1523/JNEUROSCI.3525-16.2017

Episode producers:
Karthik Sama, Xuqian Michelle Li

“Naturalistic stimuli open up new exploration…”

Dr. Christopher Baldassano is an associate professor at Columbia University and leads the Dynamic Perception and Memory Lab. With a background in electrical engineering from Princeton and a PhD in computer science from Stanford, Chris has pioneered innovative approaches to understanding memory and cognition. Following a postdoc at Princeton with Uri Hasson and Ken Norman, he joined Columbia in 2018. His research focuses on how the brain processes, stores, and retrieves events using naturalistic stimuli, hidden Markov models, and multivariate analysis techniques.

In this episode, Peter and Chris explore the fascinating world of event structures and memory. They discuss Chris’s pioneering work on event scripts, neural frameworks that act as cognitive scaffolds for autobiographical memories. The conversation covers how the brain segments continuous experience into discrete events, the role of event boundaries in memory encoding, and the critical function of the hippocampus in organizing these temporal structures. Chris explains his use of naturalistic stimuli and hidden Markov models to reveal the subtle dynamics of how we combine recurring information to respond more efficiently to future experiences. Along the way, Chris shares valuable insights on the evolution of neuroscience research and offers thoughtful advice for aspiring scientists navigating the field.

We hope you enjoy this episode!

Chapters:
00:00 - Introduction
07:37 - Transitioning from Computer Science to Neuroscience
13:01 - Exploring Naturalistic Stimuli in Neuroscience
18:11 - Hidden Markov Models in Narrative Perception
22:46 - Event Boundaries and Memory Encoding
27:49 - The Role of the Hippocampus in Memory
33:01 - Implications for Mental Health and Memory Disorders
38:19 - Enhancing Memory Techniques
41:11 - Contextualization in Memory
46:19 - Understanding Brain States
49:01 - AI and Contextual Knowledge
53:29 - Infant Cognition and Event Structures
01:01:31 - Future Directions in Research

Works mentioned:
2:28 - https://www.youtube.com/watch?v=jPLWOBmaLkY
(Baldassano talk at NIH workshop on naturalistic stimuli)
14:42 - https://pubmed.ncbi.nlm.nih.gov/28772125/
(Baldassano et al., 2017 - Neuron - "Discovering Event Structure in Continuous Narrative Perception and Memory")
15:02 - https://pubmed.ncbi.nlm.nih.gov/30249790/
(Baldassano et al., 2018 - Journal of Neuroscience - "Representation of Real-world Event Schemas During Narrative Perception")
18:24 - https://pubmed.ncbi.nlm.nih.gov/29087305/
(Vidaurre, Smith & Woolrich, 2017 - PNAS - "Brain network dynamics are hierarchically organized in time" - using Markov models in a different way)
19:41 - https://pubmed.ncbi.nlm.nih.gov/17338600/
(Zacks et al., 2007 - Psychological Bulletin - "Event perception: A mind-brain perspective" - foundational work on event boundary processes)
27:04 - https://pubmed.ncbi.nlm.nih.gov/27121839/
(Huth et al., 2016 - Nature - "Natural speech reveals the semantic maps that tile human cerebral cortex" - semantic information stored throughout the brain)
37:15 - https://pubmed.ncbi.nlm.nih.gov/22982082/
(LePort et al., 2012 - Neurobiology of Learning and Memory - Jim McGaugh's study on highly superior autobiographical memory)
53:01 - https://pubmed.ncbi.nlm.nih.gov/36252007/
(Yates et al., 2022 - PNAS - "Neural event segmentation of continuous experience in human infants")

Episode producers:
Xuqian Michelle Li

Dr. Ahmed Khalil is an MD-PhD currently serving his residency in radiology at the Institute of Neuroradiology at Charité University Hospital in Berlin. Originally from Sudan, he has been doing pioneering work on resting-state BOLD latency mapping, a technique that reveals flow deficits in the brain associated with stroke. His research demonstrates that this approach compares favorably with the current clinical gold standard of dynamic susceptibility contrast imaging using gadolinium, while capturing useful data in as little as two minutes.

In this episode, Peter and Ahmed discuss his work translating advanced MRI techniques into clinical practice. They explore how BOLD latency mapping can detect perfusion deficits and compare with both traditional gadolinium-based methods and DTI for identifying stroke lesions. The conversation delves into the broader challenge faced by all promising research methods: what does it actually take to move from successful proof-of-concept to daily clinical practice on scanners around the world?

Ahmed and Peter also talk about the cultural gap between research-level image processing and the clinical preference for minimally processed, interpretable data and how AI might help bridge that divide. Along the way, Ahmed shares valuable advice for MD-PhD students on the importance of collaboration, learning from diverse experts, and maintaining curiosity across disciplines.

We hope you enjoy this episode!

Chapters:
00:00 - Introduction to Ahmed Khalil and His Work
05:02 - Journey into Medicine and Radiology
12:10 - The Challenges of Methods Development in Clinical Applications
22:15 - Research on Resting State BOLD Latency
37:27 - Clinical Implications of Perfusion Imaging in Stroke
43:52 - Challenges in Clinical Implementation of New Imaging Techniques
47:50 - The Role of AI in Radiology and Imaging Interpretation
52:42 - Future Aspirations and Research Directions in Imaging
01:01:03 - Collaborative Efforts in Physiologic MRI Book Project
01:03:25 - Advice for Aspiring MD-PhD Students

Works mentioned:
22:48 - https://pubmed.ncbi.nlm.nih.gov/23378326/
(Lv et al., 2013 - First paper showing BOLD delay in stroke with Arno Villinger)
23:08 - https://www.ahajournals.org/doi/10.1161/STROKEAHA.116.015566
(Khalil et al., 2017 - Stroke paper, Relationship between BOLD delay and DSC-MRI)
23:08 - https://pubmed.ncbi.nlm.nih.gov/30334657/
(Khalil et al., 2018 - JCBFM paper, Longitudinal changes in BOLD delay)
39:00 - https://pubmed.ncbi.nlm.nih.gov/34323339/
(Hu et al., 2021 - Human Brain Mapping paper with Daniel Margulies - ICA approach)

Episode producers:
Ömer Faruk Gülban, Xuqian Michelle Li

“What makes certain brain networks vulnerable to disease—and can AI help us predict what comes next?”

Dr. Juan Helen Zhou is a computational neuroscientist at the National University of Singapore, where she is an Associate Professor and Director of the Center for Translational Magnetic Resonance Research at the Yong Loo Lin School of Medicine. She leads the Multimodal Neuroimaging in Neuropsychiatric Disorders Laboratory, integrating multimodal brain imaging and machine learning to study network vulnerability in aging and neuropsychiatric disorders, including dementia, psychosis, and ADHD.

In this episode, Peter and Helen discuss her path from computer science to neuroscience and how that background shaped her approach to brain imaging and AI. They explore her work on dementia, including the role of cerebral vascular disease, why different forms of dementia must be understood as distinct network-level disorders, and how selective brain network vulnerabilities can predict cognitive decline.

The discussion also covers recent advances from Dr. Zhou’s lab in reconstructing images from brain activity using generative AI and self-supervised learning, highlighting both the promise and challenges of these approaches. Along the way, Helen reflects on the importance of collaboration in neuroscience and shares advice for early-career researchers on persistence, communication, and navigating interdisciplinary science.

We hope you enjoy this episode!

Chapters:
00:00 - Introduction to Helen Zhou and Her Background
03:28 - Journey from Computer Science to Neuroscience
11:13 - The Center for Translational MR Research
12:59 - Involvement with OHBM and Community Growth
23:44 - Research Focus on Dementia and Brain Networks
28:05 - Exploring Cerebral Vasculitis and Dementia Stages
44:02 - Functional Specialization and Cognitive Performance
45:34 - AI-Based Interventions for Cognitive Health
58:30 - Utilizing Large Datasets for Brain Research
01:08:53 - Advice for Aspiring Neuroscientists

Works mentioned:
25:18 - https://www.cell.com/neuron/fulltext/S0896-6273(09)00249-9
25:18 - https://www.cell.com/neuron/fulltext/S0896-6273(12)00227-9
26:55 - https://www.neurology.org/doi/10.1212/WNL.0000000000008315
38:33 - https://www.neurology.org/doi/10.1212/wnl.0000000000207401
41:00 - https://www.sciencedirect.com/science/article/abs/pii/S1053811916002342
42:33 - https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0079419
47:46 - https://openaccess.thecvf.com/content/CVPR2023/html/Chen_Seeing_Beyond_the_Brain_Conditional_Diffusion_Model_With_Sparse_Masked_CVPR_2023_paper.html
55:11 - https://www.nature.com/articles/s41586-022-04554-y

Episode producers:
Karthik Sama, Xuqian Michelle Li

“What does it actually mean to understand the brain?”

Dr. Kendrick Kay is a computational neuroscientist and neuroimaging expert at the University of Minnesota’s Center for Magnetic Resonance Research, where he is an Associate Professor in the Department of Radiology. With training spanning philosophy and neuroscience, from a bachelor’s degree in philosophy at Harvard University to a PhD in neuroscience from UC Berkeley, Dr. Kay’s work bridges deep theoretical questions with cutting-edge neuroimaging methods.

In this conversation, Peter Bandettini and Kendrick Kay explore the evolving landscape of neuroscience at the intersection of fMRI, philosophy, and artificial intelligence. They reflect on the limits of current neuroimaging methodologies, what fMRI can and cannot tell us about brain mechanisms, and why creativity and human judgment remain central to scientific progress. The discussion also dives into Dr. Kay’s landmark contributions to fMRI decoding and the Natural Scenes Dataset, a high-resolution resource that has become foundational for computational neuroscience and neuro AI research.

Along the way, they examine deep sampling in neuroimaging, individual variability in brain data, and the challenges of separating neural signals from hemodynamic effects. Framed by broader questions about understanding benchmarking progress, and the growing role of LLM’s in neuroscience, this wide-ranging conversation offers a thoughtful look at where the field has been and where it may be headed.

We hope you enjoy this episode!

Chapters:
00:00 - Introduction to Kendrick Kay and His Work
04:51 - Philosophy’s Influence on Neuroscience
17:17 - How Far Will fMRI Take Us?
23:27 - Understanding Attention in Neuroscience
30:00 - Science as a Process
34:17 - The Role of Large Language Models (LLMs) in Scientific Progress
38:29 - Why Humans Should Stay in the Equation
40:30 - Creativity vs. AI in Scientific Research
54:48 - Dr. Kay’s Natural Scenes Dataset (NSD)
01:00:27 - Deep Sampling: Considerations and Implications
01:08:00 - Accounting for biological variation in Brain Scans: Differences and Similarities
01:13:00 - Separating Hemodynamic Effects from Neural Effects
01:16:00 - Areas of Hope and Progress in the field
01:21:00 - How Should We Benchmark Progress?
01:22:59 - Advice for Aspiring Scientists

Works mentioned:
54:48 -  https://www.nature.com/articles/s41593-021-00962-x
54:50 - https://www.sciencedirect.com/science/article/pii/S0166223624001838?via%3Dihub

Episode producers:
Xuqian Michelle Li, Naga Thovinakere