Support the show to get full episodes, full archive, and join the Discord community. Sri and Mei join me to discuss how including principles of neuromodulation in deep learning networks may improve network performance. It's an ever-present question how much detail to include in models, and we are in the early stages of learning how neuromodulators and their interactions shape biological brain function. But as we continue to learn more, Sri and Mei are interested in building "neuromodulation-aware DNNs". Neural Circuits Laboratory.Twitter: Sri: @srikipedia; Jie: @neuro_Mei.Related papersInforming deep neural networks by multiscale principles of neuromodulatory systems. 0:00 - Intro 3:10 - Background 9:19 - Bottom-up vs. top-down 14:42 - Levels of abstraction 22:46 - Biological neuromodulation 33:18 - Inventing neuromodulators 41:10 - How far along are we? 53:31 - Multiple realizability 1:09:40 -Modeling dendrites 1:15:24 - Across-species neuromodulation

Support the show to get full episodes, full archive, and join the Discord community. Eve discusses many of the lessons she has learned studying a small nervous system, the crustacean stomatogastric nervous system (STG). The STG has only about 30 neurons and its connections and neurophysiology are well-understood. Yet Eve's work has shown it functions under a remarkable diversity of conditions, and does so is a remarkable variety of ways. We discuss her work on the STG specifically, and what her work implies about trying to study much larger nervous systems, like our human brains. The Marder Lab.Twitter: @MarderLab.Related to our conversation:Understanding Brains: Details, Intuition, and Big Data.Emerging principles governing the operation of neural networks (Eve mentions this regarding "building blocks" of neural networks). 0:00 - Intro 3:58 - Background 8:00 - Levels of ambiguity 9:47 - Stomatogastric nervous system 17:13 - Structure vs. function 26:08 - Role of theory 34:56 - Technology vs. understanding 38:25 - Higher cognitive function 44:35 - Adaptability, resilience, evolution 50:23 - Climate change 56:11 - Deep learning 57:12 - Dynamical systems

Support the show to get full episodes, full archive, and join the Discord community. Patryk and I discuss his wide-ranging background working in both the neuroscience and AI worlds, and his resultant perspective on what's needed to move forward in AI, including some principles of brain processes that are more and less important. We also discuss his own work using some of those principles to help deep learning generalize to better capture how humans behave in and perceive the world. Patryk's homepage.Twitter: @paklnet.Related papersUnsupervised Learning from Continuous Video in a Scalable Predictive Recurrent Network. 0:00 - Intro 2:22 - Patryk's background 8:37 - Importance of diverse skills 16:14 - What is intelligence? 20:34 - Important brain principles 22:36 - Learning from the real world 35:09 - Language models 42:51 - AI contribution to neuroscience 48:22 - Criteria for "real" AI 53:11 - Neuroscience for AI 1:01:20 - What can we ignore about brains? 1:11:45 - Advice to past self

Support the show to get full episodes, full archive, and join the Discord community. Hakwan and I discuss many of the topics in his new book, In Consciousness we Trust: The Cognitive Neuroscience of Subjective Experience. Hakwan describes his perceptual reality monitoring theory of consciousness, which suggests consciousness may act as a systems check between our sensory perceptions and higher cognitive functions. We also discuss his latest thoughts on mental quality space and how it relates to perceptual reality monitoring. Among many other topics, we chat about the many confounds and challenges to empirically studying consciousness, a topic featured heavily in the first half of his book. Hakwan was on a previous episode with Steve Fleming, BI 099 Hakwan Lau and Steve Fleming: Neuro-AI Consciousness. Hakwan's lab: Consciousness and Metacognition Lab.Twitter: @hakwanlau.Book:In Consciousness we Trust: The Cognitive Neuroscience of Subjective Experience. 0:00 - Intro 4:37 - In Consciousness We Trust 12:19 - Too many consciousness theories? 19:26 - Philosophy and neuroscience of consciousness 29:00 - Local vs. global theories 31:20 - Perceptual reality monitoring and GANs 42:43 - Functions of consciousness 47:17 - Mental quality space 56:44 - Cognitive maps 1:06:28 - Performance capacity confounds 1:12:28 - Blindsight 1:19:11 - Philosophy vs. empirical work

Support the show to get full episodes, full archive, and join the Discord community. Tomás and I discuss his research and ideas on how memories are encoded (the engram), the role of forgetting, and the overlapping mechanisms of memory and instinct. Tomás uses otpogenetics and other techniques to label and control neurons involved in learning and memory, and has shown that forgotten memories can be restored by stimulating "engram cells" originally associated with the forgotten memory. This line of research has led Tomás to think forgetting might be a learning mechanism itself, a adaption our brains make based on the predictability and affordances of the environment. His work on engrams has also led Tomás to think our instincts (ingrams) may share the same mechanism of our memories (engrams), and that memories may transition to instincts across generations. We begin by addressing Randy Gallistel's engram ideas from the previous episode: BI 126 Randy Gallistel: Where Is the Engram? Ryan Lab.Twitter: @TJRyan_77.Related papersEngram cell connectivity: an evolving substrate for information storage.Forgetting as a form of adaptive engram cell plasticity.Memory and Instinct as a Continuum of Information Storage in The Cognitive Neurosciences.The Bandwagon by Claude Shannon. 0:00 - Intro 4:05 - Response to Randy Gallistel 10:45 - Computation in the brain 14:52 - Instinct and memory 19:37 - Dynamics of memory 21:55 - Wiring vs. connection strength plasticity 24:16 - Changing one's mind 33:09 - Optogenetics and memory experiments 47:24 - Forgetting as learning 1:06:35 - Folk psychological terms 1:08:49 - Memory becoming instinct 1:21:49 - Instinct across the lifetime 1:25:52 - Boundaries of memories 1:28:52 - Subjective experience of memory 1:31:58 - Interdisciplinary research 1:37:32 - Communicating science

Support the show to get full episodes, full archive, and join the Discord community. Randy and I discuss his long-standing interest in how the brain stores information to compute. That is, where is the engram, the physical trace of memory in the brain? Modern neuroscience is dominated by the view that memories are stored among synaptic connections in populations of neurons. Randy believes a more reasonable and reliable way to store abstract symbols, like numbers, is to write them into code within individual neurons. Thus, the spiking code, whatever it is, functions to write and read memories into and out of intracellular substrates, like polynucleotides (DNA, RNA, e.g.). He lays out his case in detail in his book with Adam King, Memory and the Computational Brain: Why Cognitive Science will Transform Neuroscience. We also talk about some research and theoretical work since then that support his views. Randy's Rutger's website.Book:Memory and the Computational Brain: Why Cognitive Science will Transform Neuroscience.Related papers:The theoretical RNA paper Randy mentions: An RNA-based theory of natural universal computation.Evidence for intracellular engram in cerebellum: Memory trace and timing mechanism localized to cerebellar Purkinje cells.The exchange between Randy and John Lisman.The blog post Randy mentions about Universal function approximation:The Truth About the [Not So] Universal Approximation Theorem 0:00 - Intro 6:50 - Cognitive science vs. computational neuroscience 13:23 - Brain as computing device 15:45 - Noam Chomsky's influence 17:58 - Memory must be stored within cells 30:58 - Theoretical support for the idea 34:15 - Cerebellum evidence supporting the idea 40:56 - What is the write mechanism? 51:11 - Thoughts on deep learning 1:00:02 - Multiple memory mechanisms? 1:10:56 - The role of plasticity 1:12:06 - Trying to convince molecular biologists

Support the show to get full episodes, full archive, and join the Discord community. Doris, Tony, and Blake are the organizers for this year's NAISys conference, From Neuroscience to Artificially Intelligent Systems (NAISys), at Cold Spring Harbor. We discuss the conference itself, some history of the neuroscience and AI interface, their current research interests, and a handful of topics around evolution, innateness, development, learning, and the current and future prospects for using neuroscience to inspire new ideas in artificial intelligence. From Neuroscience to Artificially Intelligent Systems (NAISys).Doris:@doristsao.Tsao Lab.Unsupervised deep learning identifies semantic disentanglement in single inferotemporal face patch neurons.Tony:@TonyZador.Zador Lab.A Critique of Pure Learning: What Artificial Neural Networks can Learn from Animal Brains.Blake:@tyrell_turing.The Learning in Neural Circuits Lab.The functional specialization of visual cortex emerges from training parallel pathways with self-supervised predictive learning. 0:00 - Intro 4:16 - Tony Zador 5:38 - Doris Tsao 10:44 - Blake Richards 15:46 - Deductive, inductive, abductive inference 16:32 - NAISys 33:09 - Evolution, development, learning 38:23 - Learning: plasticity vs. dynamical structures 54:13 - Different kinds of understanding 1:03:05 - Do we understand evolution well enough? 1:04:03 - Neuro-AI fad? 1:06:26 - Are your problems bigger or smaller now?

Support the show to get full episodes, full archive, and join the Discord community. Robin and I discuss many of the ideas in his book The Self-Assembling Brain: How Neural Networks Grow Smarter. The premise is that our DNA encodes an algorithmic growth process that unfolds information via time and energy, resulting in a connected neural network (our brains!) imbued with vast amounts of information from the "start". This contrasts with modern deep learning networks, which start with minimal initial information in their connectivity, and instead rely almost solely on learning to gain their function. Robin suggests we won't be able to create anything with close to human-like intelligence unless we build in an algorithmic growth process and an evolutionary selection process to create artificial networks. Hiesinger Neurogenetics LaboratoryTwitter: @HiesingerLab.Book: The Self-Assembling Brain: How Neural Networks Grow Smarter 0:00 - Intro 3:01 - The Self-Assembling Brain 21:14 - Including growth in networks 27:52 - Information unfolding and algorithmic growth 31:27 - Cellular automata 40:43 - Learning as a continuum of growth 45:01 - Robustness, autonomous agents 49:11 - Metabolism vs. connectivity 58:00 - Feedback at all levels 1:05:32 - Generality vs. specificity 1:10:36 - Whole brain emulation 1:20:38 - Changing view of intelligence 1:26:34 - Popular and wrong vs. unknown and right

Support the show to get full episodes, full archive, and join the Discord community. Irina is a faculty member at MILA-Quebec AI Institute and a professor at Université de Montréal. She has worked from both ends of the neuroscience/AI interface, using AI for neuroscience applications, and using neural principles to help improve AI. We discuss her work on biologically-plausible alternatives to back-propagation, using "auxiliary variables" in addition to the normal connection weight updates. We also discuss the world of lifelong learning, which seeks to train networks in an online manner to improve on any tasks as they are introduced. Catastrophic forgetting is an obstacle in modern deep learning, where a network forgets old tasks when it is trained on new tasks. Lifelong learning strategies, like continual learning, transfer learning, and meta-learning seek to overcome catastrophic forgetting, and we talk about some of the inspirations from neuroscience being used to help lifelong learning in networks. Irina's website.Twitter: @irinarishRelated papers:Beyond Backprop: Online Alternating Minimization with Auxiliary Variables.Towards Continual Reinforcement Learning: A Review and Perspectives.Lifelong learning video tutorial: DLRL Summer School 2021 - Lifelong Learning - Irina Rish. 0:00 - Intro 3:26 - AI for Neuro, Neuro for AI 14:59 - Utility of philosophy 20:51 - Artificial general intelligence 24:34 - Back-propagation alternatives 35:10 - Inductive bias vs. scaling generic architectures 45:51 - Continual learning 59:54 - Neuro-inspired continual learning 1:06:57 - Learning trajectories

Support the show to get full episodes and join the Discord community. Ko and I discuss a range of topics around his work to understand our visual intelligence. Ko was a postdoc in James Dicarlo's lab, where he helped develop the convolutional neural network models that have become the standard for explaining core object recognition. He is starting his own lab at York University, where he will continue to expand and refine the models, adding important biological details and incorporating models for brain areas outside the ventral visual stream. He will also continue recording neural activity, and performing perturbation studies to better understand the networks involved in our visual cognition. VISUAL INTELLIGENCE AND TECHNOLOGICAL ADVANCES LABTwitter: @KohitijKar.Related papersEvidence that recurrent circuits are critical to the ventral stream’s execution of core object recognition behavior.Neural population control via deep image synthesis.BI 075 Jim DiCarlo: Reverse Engineering Vision 0:00 - Intro 3:49 - Background 13:51 - Where are we in understanding vision? 19:46 - Benchmarks 21:21 - Falsifying models 23:19 - Modeling vs. experiment speed 29:26 - Simple vs complex models 35:34 - Dorsal visual stream and deep learning 44:10 - Modularity and brain area roles 50:58 - Chemogenetic perturbation, DREADDs 57:10 - Future lab vision, clinical applications 1:03:55 - Controlling visual neurons via image synthesis 1:12:14 - Is it enough to study nonhuman animals? 1:18:55 - Neuro/AI intersection 1:26:54 - What is intelligence?