Neuroscience Tutorial

Brain-AI Workshop 2017

This document serves as an outline of some of the important differences between the mammalian brain and artificial neural networks.

Spikes vs. Rate

Brains are stochastic

Vanishing/exploding gradients / criticality

Brains have multiple conflicting objective functions

Brains have a continuum of timescales

Brains are modular recurrent networks

Term Meaning Module
Episodic Memory Autobiographical memory. E.g. what you had for breakfast yesterday. Where you went to school. medial temporal lobe / acetylcholine
Source Memory Knowledge about how you know something. E.g. when did you first hear about action potentials?
Semantic Memory Information about the world. The president of China. The first month of the year. distributed?
Procedural Memory How to do something. E.g. play the piano, ride a bike. motor cortex / basal ganglia / dopamine
Emotional memory An association between a thing and a value (good or bad). Chocolate tastes good. Fire burns. A certain dog bites. Amygdala
Working memory a short-term form of memory often associated with combining or integrating multiple items in memory together. E.g. doing math problem in your head, trying to count cards at a casino, finding the largest number in a list. Generally, the items in memory are discarded soon after finishing. prefrontal cortex (for longer durations, hippocampus).
Spatial working memory like working memory but in particular for a location in space. e.g. when playing basketball keeping track of where the players are on the court. frontal eye field and other regions.
LTP/D Long-term potentiation/depression. Changes in synaptic strength that last for more than 1 day. Requires new protein synthesis and changes to cyto-skeleton
STP/D Short-term potentiation/depression. Changes in synpatic strength that last for a short time (usually 30 min - a few hours)
Spatial Attention enhancing processing of one part of the world. frontal-parietal-tectal circuit, acetylcholine
Feature Attention Focused processing on a non-spatial dimension of the world. E.g. looking for a friend in a red jacket. Focused listening of one instrument at the symphony. Depends on modality. Higher sensory cortex with PFC
Overt vs. Covert attention Overt attention is externally observable. e.g. by measuring eye/head/body positions. Covert attention is not-externally observable. e.g. "looking" at someone out of the corner of your eye or shifting your focus from vision to hearing or touch. Depends. For switching modulality, probably thalamus.
Habit an action that is performed regardless of the outcome. For example, if you eat a snack every day at 5pm, and then one day you have a huge meal at 4pm and you are really full but you still eat the snack, then it is a habit.
Goal-directed action opposite of habit. A behavior that is performed to achieve a goal. Generally more flexible than habits. dopamine
Motivation Used to describe levels of drive. (i.e. scaling of objective functions). When we are thirsty we are motivated to get water. But once we are not thirsty the same outcome (water) has less value. dopamine, noradrenaline, opiates
Behavioral inhibition e.g. self-control Ability to suppress actions based on context. E.g. you want to eat the cookie, but you are trying to lose weight, or you want your friend to have the last one prefrontal cortex, serotonin
perceptual decision-making Making a decision about some physical property of an object. E.g. is that picture hanging straight? Is there a pothole in the road? Will the ping-pong ball miss the table probably secondary sensory cortex.
value based decision-making also called economic or financial decision-making. Should I buy this stock? Should I ask that girl out on a date? Should I wait for the bus? Most real decisions involve both perceptual and value-based parts. striatum / nucleus accumbens / orbital frontal cortex / dopamine


The brain is very plastic (meaning the connection weights can be changed). There is a huge amount of diversity in the learning rules in the brain. For almost every cognitive function listed above there is a corresponding learning rule.

Why have such diverse rules? Probably to match statistics of environment. Things that don't change much should have correspondingly slow learning rates.

We know a lot about the molecular mechanisms of learning. Probably the best studied feature of the brain.

Local Learning

Global Learning

Supervised Learning

Unsupervised learning

Perceptual/Statistical Learning == autoencoder

Motor Learning

hierarchical learning

Are modern recurrent networks (LSTM/GRU) biologically plausible?

Why should be care about any of this?

  1. see 21-rate.pdf 

  2. Kopec, Charles D, Jeffrey C Erlich, Bingni W Brunton, Karl Deisseroth, and Carlos D Brody. “Cortical and Subcortical Contributions to Short- Term Memory for Orienting Movements.” Neuron 88, no. 2 (October 2015): 367–377. doi:10.1016/j.neuron.2015.08.033.  

    Wang, Xiao-Jing. “Decision Making in Recurrent Neuronal Circuits.” Neuron 60, no. 2 (October 23, 2008): 215–34. doi:10.1016/j.neuron.2008.09.034.

    Wong, Kong-Fatt, and Xiao-Jing Wang. “A Recurrent Network Mechanism of Time Integration in Perceptual Decisions.” Journal of Neuroscience 26, no. 4 (January 25, 2006): 1314–28. doi:10.1523/JNEUROSCI.3733-05.2006.

  3. Rate vs. temporal coding 

  4. Dean HL, Hagan MA, Pesaran B. Only coherent spiking in posterior parietal cortex coordinates looking and reaching. Neuron. 2012. 73(4):829-841) 

  5. Delcomyn, F. “Neural Basis of Rhythmic Behavior in Animals.” Science 210, no. 4469 (October 31, 1980): 492–98. doi:10.1126/science.7423199. 

    Marder, Eve, and Dirk Bucher. “Central Pattern Generators and the Control of Rhythmic Movements.” Current Biology 11, no. 23 (November 27, 2001): R986–96. doi:10.1016/S0960-9822(01)00581-4.


  7. Maynard Smith J. 1982. Evolution and the Theory of Games. London: Cambridge Univ. Press 


  9. Egorov, Alexei V., Bassam N. Hamam, Erik Fransén, Michael E. Hasselmo, and Angel A. Alonso. “Graded Persistent Activity in Entorhinal Cortex Neurons.” Nature 420, no. 6912 (November 14, 2002): 173–78. doi:10.1038/nature01171.