July 13-31, 2020
Objectives: Introduce traditional and emerging computational neuroscience tools, their complementarity, and what they can tell us about the brain. A main focus is on modeling choices, model creation, model evaluation and understanding how they relate to biological questions.
Tutorial microstructure: ~10min talk, ~20min tutorial (repeated)
Day structure: Opening keynote, 3h lecture/tutorial modules, 1h interpretation (what did we learn today, what does it mean, underlying philosophy, 1h professional development/ meta-science, evening group projects (for interactive track). There will also be many networking activities!
Prerequisites: See here
Introduction of datasets (spikes, EEG, fMRI + behavior), and questions about them. These questions will foreshadow the whole summer school.
Intro / keynote & tutorial setup (0:00 - 0:50): NMA organization, expectations, code of conduct, modeling vs. data
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Data intro, preprocessing | Link of neural data to behavior | Tuning (RFs, motor, STA) | What it means to "understand" (signal detection) |
Recap, Q&A (4:35 - 5:30): Outlook on school
Professional development (5:30 - 6:00): Being a good NMA participant
Introduce different example model types (Marr 1-3, what/how/why) and the kinds of questions they can answer. MRealize how different models map onto different datasets.
Intro / keynote & tutorial setup (0:00 - 0:50): Model classifications
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Marr 1 | Marr 2-3 | "What" | "How"/"Why" |
Recap, Q&A (4:35 - 5:30): The role of models in discovery
Professional development (5:30 - 6:00): How-to-model guide 1
Fit models to data, quantify uncertainty, compare models
Intro / keynote & tutorial setup (0:00 - 0:50): Why and how to fit models
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Fit a model 1 (linear regression) | Get error bars | Compare models, cross-validation, hyperparameters | Fit a model 2 (nonlinear models) |
Recap, Q&A (4:35 - 5:30): Critical evaluation of model fitting
Professional development (5:30 - 6:00): How-to-model guide 2
Introduction to machine learning. The commonly used approaches, how to avoid false positives, how to do it well
Intro / keynote & tutorial setup (0:00 - 0:50): We want to predict (scikitlearn)...
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| GLMs (temporal filtering models) | Linear classifier (SVM) | Regularization (L1, L2) | Shallow nonlinear classifier (SVM with RBF kernel) |
Recap, Q&A (4:35 - 5:30): Promises and pitfalls of ML
Professional development (5:30 - 6:00): How-to-model guide 3
Concept of dimensionality reduction, ways of doing it, what it means
Intro / keynote & tutorial setup (0:00 - 0:50): Manifolds to understand
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| PCA 1 | PCA 2 (+CCA/clustering) | Signal vs. noise manifolds | Visualizing high-D nonlinear manifolds (e.g. tSNE) |
Recap, Q&A (4:35 - 5:30): The link between high-dimensional brain signals and low-dimensional behavior
Professional development (5:30 - 6:00): Efficient collaborations
Bayesian statistics, modeling of behavior, modeling of neural data, quantifying information
Intro / keynote & tutorial setup (0:00 - 0:50): Uncertainty
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Bayes rule I (product rule: cue combination) | Bayes rule II (Marginalization and nuisance variables) | Causal inference & structural models (use as example for marginalization) | Bayesian decision theory |
Recap, Q&A (4:35 - 5:30): Advanced Bayesian methods
Professional development (5:30 - 6:00): Productivity tools for science
How to make estimates over time, how the brain does it
Intro / keynote & tutorial setup (0:00 - 0:50): World has time
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Linear systems theory I (ND deterministic) | Linear systems theory II (1D stochastic = OU process; ND stocastic = AR(1)) | Markov process | State space model |
Recap, Q&A (4:35 - 5:30): Linear systems rule the world
Professional development (5:30 - 6:00): Open source ecosystem, data management & sharing
How we can make decisions when information comes in over time
Intro / keynote & tutorial setup (0:00 - 0:50): We need to decide stuff
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Information theory | Sequential Probability Ratio Test (SPRT) | Hidden Markov Model inference (DDM) | Kalman filter |
Recap, Q&A (4:35 - 5:30): Decisions, decisions, decisions...
Professional development (5:30 - 6:00): Open science (general), replicability & reproducibility
We need to move gain info and reach goals
Intro / keynote & tutorial setup (0:00 - 0:50): We want to control our actions...
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Expected utility / cost | Markov decision process (MDP) | LQG control (MDP for linear systems) | Motor control (signal-dependent noise, time cost, ...) |
Recap, Q&A (4:35 - 5:30): Advanced motor control
Professional development (5:30 - 6:00): Networking at Conferences
The setting of reinforcement learning and how it approximates the real world, behavior, and potential brain implementations
Intro / keynote & tutorial setup (0:00 - 0:50): Problem formulations: actor-critic
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Critic: reward prediction error | Exploration (POMDP) vs exploitation | Model-based vs model-free RL | Multi-arm bandits: foraging |
Recap, Q&A (4:35 - 5:30): RL and the brain
Professional development (5:30 - 6:00): Writing Papers & Grants
The things neurons are made of, channels, morphologies, neuromodulators, and plasticity
Intro / keynote & tutorial setup (0:00 - 0:50): Real neurons ftw
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Channels, HH | LIF neuron | LNP (loses fine timing info) | Hebbian plasticity & STDP |
Recap, Q&A (4:35 - 5:30): A variety of neuron models
Professional development (5:30 - 6:00): How to find a postdoc
How single neurons create population dynamics
Intro / keynote & tutorial setup (0:00 - 0:50): Mechanistic models of different types of brain actvivity
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Spikes to rates | Wilson-Cowen model (coarse-grained), oscillations & synchrony | Attractors & local linearization around fixed points | Chaos in rate networks (stimulus dependent chaotic attractor) |
Recap, Q&A (4:35 - 5:30): A theory of the whole brain
Professional development (5:30 - 6:00): Early career panel - academia (how to advance through career steps)
Ways of discovering causal relations, ways of estimating networks, what we can do with networks
Intro / keynote & tutorial setup (0:00 - 0:50): Causality - different views
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Pitfalls of Granger | Centrality | Instrumental variables | Interventions |
Recap, Q&A (4:35 - 5:30): Latters of causality
Professional development (5:30 - 6:00): Computational neuroscience in industry - career panel
The concept of ANNs, how to train them,what they are made out of, convnets, and how to fit them to brains
Intro / keynote & tutorial setup (0:00 - 0:50): DL = crucial tool
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Pytorch intro & model components | Training it & inductive bias | ConvNets | Fit to brain (RSA - represenatational similarity analysis) |
Recap, Q&A (4:35 - 5:30): Digging deep
Professional development (5:30 - 6:00): Job fair (FRL)
Deep learning in more advanced settings. Autoencoders for structure discovery, RNNs, and fitting them to brains
Intro / keynote & tutorial setup (0:00 - 0:50): DL for structure
| Lecture & tutorial 1 | Lecture & tutorial 2 | Lecture & tutorial 3 | Lecture & tutorial 4 |
|---|---|---|---|
| 0:50 - 1:25 | 1:30 - 2:05 | 2:10 - 2:45 | 3:30 - 4:05 |
| Autoencoders | RNN | Transfer learning / generalization | Causality |
Recap, Q&A (4:35 - 5:30): Digging deeper
Professional development (5:30 - 6:00): NMA wrap-up
- Meet a prof about your group's project
- Meet a prof about your career
- Meet a prof about your own project
- Meet other participants interested in similar topics
- Meet a group of likeminded people
- Meet people that are local to you (same city, country)
TBA