The MBC/IGERT Graduate Training Program

One of the main goals of the Center for Mind, Brain, and Computation is to train future scientists to investigate the emergent functions of neural systems in the brain using a combination of computational and experimental approaches. This goal is addressed by the MBC/IGERT graduate training program. The program seeks outstanding students who will combine rigorous mathematical and computational methods with experimental investigations to address emergent functions of neural systems. The program is supported by an NSF Integrative Graduate Education and Research Training (IGERT) Grant.

Our program is grounded in the idea that contemporary research on the emergent functions of the nervous system often depends on the synergistic integration of experimental and computational methods. In seeking to train students to integrate these methods, a flexible, learn-through-experience approach is needed. The program is built around a flexible program of individually-chosen courses that will allow each student to gain just the right background for their needs, followed by the MBC Research Experience, which generally will exploit the specific background acquired through the coursework. The program also seeks to provide a supportive overall training environment through course offerings, seminars, and other events that will help foster fuller integration of computational/quantitative and experimental approaches at many levels of investigation in neuroscience.

The program seeks especially to train students who wish to make a serious commitment to extend their research training and research skill set in the direction of achieving an integration of approaches. Two levels of participants in the program are recognized: Trainees and Affiliates.

Trainees are Ph. D. students who commit to a specialized training program, including individually selected course work and an integrative research project, as discussed below. Trainees who are US nationals are eligible to be considered for two years of stipend and partial tuition support funded by the IGERT grant. Affiliates are Ph.D. students or other members of the Stanford community who are interested in the research themes and approach of MBC, but who are either (a) preparing to become trainees or (b) benefiting from the activities of the program without making the full commitment required of trainees. For information on Joining MBC either as an affiliate or as a trainee, see Join MBC.

Program Structure

Students who are admitted to a participating Ph.D. program (Computer Science, Electrical Engineering, Neurobiology, or Psychology) may concurrently or subsequently apply to become affiliates of MBC. Affiliate status is a necessary first step toward becoming a trainee.

Prospective trainees create an individualized training program, comprising a series of courses that provide a strong grounding in a new research method complementing the student’s primary Ph. D. training. Integrative educational experiences, including the MBC Research Experience, accompany that specialized training. The MBC Research Experience allows students to utilize their new knowledge under the joint supervision of a primary home-department mentor and a secondary mentor with appropriately balanced expertise.

Other program activities

The MBC offers ongoing weekly seminar series, largely devoted to student research presentations and coordinated by advanced program students working under the supervision of a member of the training faculty. All trainees are expected to attend on a regular basis, along with at least two faculty chosen by the presenting student. Special sessions throughout the year are used for orientation of new program participants, presentation of specialized ethics, professional development and responsible conduct of research presentations, and for open discussion of the state of the program to allow student input to shape the program’s evolution.

The MBC also holds an annual retreat, bringing the training faculty and student participants together. Sessions with student and faculty speakers will be organized to highlight newly emerging methods and topics of broad interest to the trainees in the program.

Finally, students in the program (with input from a faculty mentor) will organize and coordinate an outside speaker series. Speakers will be invited to spend two days on campus presenting their work and interacting with students. Two such speakers will be invited each year.

Individualized Training Plan and Mentoring

Each student is expected to formulate an individualized training and research plan (ITP). One goal of this plan is to prepare the student for the MBC research experience that complements or extends the training provided by the primary research department and mentor.

The ITP will generally involve five courses, three of which should go beyond the requirements of the student’s home department. For example, a student admitted to the Neuroscience Graduate Program (NGP) might take one computer science course, one robotics course, and one psychology course (outside NGP requirements), and might select two computational neuroscience courses from available NGP program options (including new course offerings as described below, which will be included as NGP options).

The program leadership will work with each student and the student’s primary advisor to identify a secondary faculty mentor to play a key role in the student’s education and research training.

MBC Research Experience

A central element of the Stanford MBC program is the MBC Research Experience, in which students integrate both computational and experimental expertise through research supervised by both mentors.

The exact structure of the MBC research experience will vary from case to case. A student might conduct a multiple-quarter research project in the secondary mentor’s laboratory, acquiring expertise in a method not available in the primary mentor’s lab. Alternatively, the student might also divide time between the two labs for an extended period, or import a method from the secondary lab into activities within the primary laboratory, under ongoing joint supervision of both mentors. For further details, see Information for Potential Trainees.

Ethics, Professional Development, and Responsible Conduct of Research

Several sessions in the weekly research seminar each year are devoted to the topics of ethics and professional responsibility. Career “survival skills” topics include research collaboration, grant writing, graphic presentation of data, oral presentation, teaching effectively, getting a job, and alternative career paths such as biotechnology.

Another important seminar topic is the discussion of guidelines for responsible conduct of research. This provides a context for students from each discipline to gain exposure to unfamiliar ethical issues relevant to the other contributing disciplines.

Special topics that arise in the context of research collaboration between disciplines include discussion of different norms of publication and peer review. Particular emphasis will be placed on team research issues such as authorship of collaborative papers, data sharing between labs, and cultural differences in research practices between disciplines.

Relevant Courses

This list is currently under development, and new courses will be added soon.

NBIO 206.The Nervous System

Structure and function, including neuroanatomy, neurophysiology, and systems neurobiology. Topics include the properties of neurons and the mechanisms and organization underlying higher functions. Framework for general work in neurology, neuropathology, clinical medicine, and for more advanced work in neurobiology. Lecture and lab components must be taken together. 7 - 8 units, R. Dolmetsch.

Offered Winter 2008-2009, MTRF 10:00AM - 11:00 AM, R 1:15PM, location TBA.

PSYCH 209a. The Neural Basis of Cognition: A Parallel Distributed Processing Approach

Models and data to support the notion that brain representations are patterns of activity over widely dispersed populations of neurons, that mental processing involves coherent distributed engagement of neurons in these populations, and that learning and development occur primarily through the adjustment of the strengths of the connections between the neurons. How models may be used to explain aspects of human cognition, development, and effects of brain damage on cognition. Prerequisites: linear algebra, differential equations, a programming course, and two courses in psychology or neuroscience. 4 units, J. McClelland.

Offered Winter 2008-2009, MWF 11:00AM - 12:15PM, 420-417.

CS 228. Structured Probabilistic Models: Principles and Techniques

Probabilistic modeling languages for representing complex domains, algorithms for reasoning and decision making using these representations, and learning these representations from data. Focus is on probabilistic graphic models, including Bayesian and Markov networks, extensions to temporal modeling such as hidden Markov models and dynamic Bayesian networks, and extensions to decision making such as influence diagrams. Basic techniques and their applications to domains including speech recognition, biological modeling and discovery, medical diagnosis, message encoding, vision, and robot motion planning. Prerequisites: basic probability theory and algorithm design and analysis. 3 units, D. Koller.

Offered Winter 2008-2009, TR 11:00AM - 12:15PM, location TBA.

CS 229. Machine Learning

Topics: statistical pattern recognition, linear and non-linear regression, non-parametric methods, exponential family, GLIMs, support vector machines, kernel methods, model/feature selection, learning theory, VC dimension, clustering, density estimation, EM, dimensionality reduction, ICA, PCA, reinforcement learning and adaptive control, Markov decision processes, approximate dynamic programming, and policy search. Prerequisites: linear algebra, and basic probability and statistics. 3 units, A. Ng.

Offered Fall 2008-2009, MW 9:30AM - 10:45AM, Gates B1.

BIOE 341. Computational Neural Networks

Distributed neural network implementations of algorithms for signal processing, function approximation, and control. Representation of information in networks of spiking neurons. Supervised and unsupervised learning algorithms. Radial basis functions, principal and independent components analysis, reinforcement learning, support-vector machines, self-organizing maps, auto-associative learning, hidden Markov models. Related methods from information theory, signal processing, bayesian estimation, and stochastic systems. Final project in software or programmable hardware. Prerequisites: linear algebra, dynamic systems, and probability theory as in MATH 103, EE 102A, and EE 178 or equivalent, and programming experience in C++ or Matlab. 3 units, T. Sanger.

Offered Fall 2008-2009, MW 12:35PM - 2:05PM, Herrin T195.

NENS 220. Computational Neuroscience: Cells and Circuits

The course focuses on the application of computational approaches to better understand the roles of individual molecules in the behavior of neurons, circuits, and organisms. Questions addressed include: How does the membrane trafficking of particular ion channels to different neuronal regions (axons, dendrites, cell bodies) affect the computation performed by individual neurons? How is the dynamic range of neurons extended by the expression of a variety of ion channels? Are neurons analog or digital devices, or both? How can the deletions of rare forms of ion channel subunits result in complex behavioral phenotypes? What are the roles of specific neurons and circuits in those phenotypes?

The course begins with a discussion of the roles of various intrinsic excitability mechanisms (voltage-gated and leak ion channels, membrane transporters, ligand gated channels, etc.) in determining the overall resting state of neurons and their input/output relationships. The course then continues to explore microcircuit behavior and the development of emergent network behaviors. We will use the NEURON environment to illustrate the utility of simulation for understanding both cellular and network behavior, as well as for generating falsifiable hypotheses that can be tested in relevant biological systems. A final project will be developed based on the fundamentals and tools introduced in the course. Recommended: Neurobiology 206 and facility with linear algebra and calculus. 4 units, Offered Winter quarter in odd numbered years Recommended for all Neuroscience Program graduate students; open to graduate, medical, and advanced undergraduate students (with consent of the instructor). 4 units, J. Huguenard.

Offered Winter 2008-2009, MR 2:30PM - 4:15PM, H3150, Neurology Conference Room, Boswell Building, Stanford Hospital. See http://nens220.stanford.edu.

Courses in Visualization of the Nervous System

Over the last several years there has been dramatic growth and development of methods for visualization in the nervous system, and in concert with this two whole year-long sequences of courses have arisen at Stanford, addressing very different levels of analysis.

At the macroscopic level, a sequence of three courses is taught by Professors Brian Wandell (Chair, Psychology), Gary Glover (Director of the Lucas Center for Imaging), and Kalanit Grill-Spector (Professor, Psychology). This sequence describes the physics of magnetic resonance, different ways to control magnetic resonance imagers to measure chemical properties, diffusion, and functional activity; methods for measuring animal and human brain structure and activity using magnetic resonance; experimental designs, statistical and signal processing methods; and modeling from cellular signals to BOLD. This sequence introduces our students to the relationship between measurements at the level of functional magnetic resonance imaging (fMRI) and cellular signals.

A parallel sequence of courses, addressing visualization at a micro level, is taught by Professors Stephen Block (Biological Sciences), Mark Schnitzer (Applied Physics and Biological Sciences) and Stephen Smith (Molecular and Cellular Physiology). In this course students learn a variety of imaging methods that span multiple length scales. This two-course sequence explains microscope optics, resolution limits, single-molecule fluorescence, FRET, confocal microscopy, two-photon microscopy, and optical trapping.

Course Under Development

The course listed below is under development and may be offered during 2009-2010:

Electrophysiological Neural Data Analysis

Topics to be covered include an introduction to massively-parallel electrode array technology and signal acquisition; in-depth coverage of statistical signal processing, spectral estimation methods, state space methods, latent variable models, dimensionality reduction methods (e.g., PCA, FA, ISOMAP), linear and non-linear dynamical systems, convex optimization, linear-nonlinear-probabilistic models, information theory, and machine learning approaches (e.g., EM) in handling both single-channel and multi-channel point process data (action potentials) and continuous data (local field potentials).

In addition to pedagogical lectures on the basic analytic techniques, we will structure homework problem sets around the analysis of real neural data and employ MATLAB toolboxes. Lectures will introduce fundamental concepts, while discussion groups will focus on relevant literature examples of proper application of these techniques. A final project will be developed based on the fundamentals and tools introduced in the course. Prerequisites are calculus, linear algebra, and multivariate statistics or signal processing.

Krishna Shenoy (Electrical Engineering) will serve as course director, and Bill Newsome (Chair, Neurobiology and HHMI), Stephen Baccus (Neurobiology) and Terrence Sanger (Neurology) are among those who will contribute to this course offering.

Training Program Steering Committee

Laura Hope (Program Administrator)