Theoretical Neuroscience Podcast

Join Paul Middlebrooks, the producer of the BrainInspired podcast, alongside neuro-AI workshop organizer Mikkel Lepperød, and distinguished neuroscientists Ken Harris and Andreas Tolias. They explore the fascinating intersection of AI and neuroscience while cruising the scenic Norwegian fjords. Topics include the transformative impact of AI on research methodologies, the nuances of model interpretability, and the importance of collaboration in scientific inquiry. They also ponder the future implications of AI in understanding the brain and consciousness.

Most of what we have learned about the functioning of the living brain has come from extracellular electrical recordings, like the measurement of spikes, LFP, ECoG and EEG signals. And most analysis of these recordings has been statistical, looking for correlations between the recorded signals and what the animal/human is doing or being exposed to.   However, starting with the neuron rather than the data, these electrical brain signals can also be computed from biophysics-based forward models, and this is topic of this podcast. 

The most prominent visual characteristic of neurons is their dendrites. Even more than 100 years after their first observation by Cajal, their function is not fully understood. Biophysical modeling based on cable theory is a key research tool for exploring putative functions, and today’s guest is one the leading researchers in this field.   We talk about of passive and active dendrites, the kind of filtering of synaptic inputs they support, the key role of synapse placements, and how the inclusion of dendrites may facilitate AI.    

Christof Koch, a trailblazer in consciousness studies, dives deep into the enigma of why groups of atoms can feel, sparking a fascinating discussion. He explores Integrated Information Theory and its implications for understanding consciousness. The conversation also navigates the neural basis of face recognition and the contrasting states of consciousness. Koch emphasizes advancements in measuring consciousness and the intriguing connections between psychedelics and perception. Finally, he ponders the ethical dimensions of consciousness in relation to food choices and the potential of artificial consciousness.

Computational neuroscientists use many software tools, and NEURON has become the leading tool for biophysical modeling of neurons and neural network. Today’s guest has been the leading developer of NEURON since the infancy almost 50 years ago. We talk about how the tool got started and the development up until today’s modern version of the software, including CoreNEURON optimized for parallel execution of large-scale network models on multicore supercomputers.

The idea that memories are stored in molecules was popular in the middle of the 20th century. However, since the discovery of long-term potentiation (LTP) in the 1970s, the dominant view has been that our memories are stored in synapses, that is, in the connections between neurons. Today, there are signs that the interest in molecular memory is returning, and the guest has presented a theory suggesting that molecular and synaptic memory might serve complementary needs for animals. 

Is quantum physics important in determining how living systems, including brains, work? Today's guest is a professor of molecular genetics at the University of Surrey in England and explores this question in the book “Life at the edge: The coming of age of quantum biology”. In this “vintage” episode, recorded in late 2019, we talk about how quantum physics is or may be key in photosynthesis, smelling, navigation, evolution and even thinking. And we also touch on development of new antibiotics, another expertise of McFadden.

Most computational neuroscientists investigate electric dynamics in neurons or neural networks, but there is also computations going on inside neurons. Here the key dynamical variables are concentrations of numerous different molecules, and the signaling is typically done in cascades of chemical reactions, called signaling pathways. Today’s guest is an expert in this kind of modelling and is particularly interested in the signaling role of calcium.

Today’s AI is largely based on supervised learning of neural networks using the backpropagation-of-error synaptic learning rule. This learning rule relies on differentiation of continuous activation functions and is thus not directly applicable to spiking neurons. Today’s guest has developed the algorithm SuperSpike to address the problem. He has also recently developed a biologically more plausible learning rule based on self-supervised learning. We talk about both.  

Over the last ten years or so, the MindScope project at the Allen Institute in Seattle has pursued an industrylab-like approach to study the mouse visual cortex in unprecedented detail using electrophysiology, optophysiology, optical imaging and electron microscopy.  Together with collaborators at Allen, today’s guest has worked to integrate of these data into large-scale neural network, and in the podcast he talks about their ambitious endeavor.