PAST EVENTS

Connecting natural and artificial neural networks with functional brain imaging

Fri, May 15, 2015

Center for Law and Justice
CLJ 123 Washington Street ROOM 572

Abstract: Functional neuroimaging and neural network modeling were both introduced to cognitive science in the 1980s, and have both produced influential research. Yet surprisingly, the programs have advanced with little mutual influence. I will describe two different approaches to more directly connecting cognitive neural network models with functional brain imaging in the search for the neural bases of cognition. In the first, direct estimates of gross neural connectivity in real brains are used to shape the architecture of an artificial neural network (ANN).

SPEAKERS:
Timothy Rogers

Simulations with the resulting model are then used to understand patterns of healthy and impaired behaviors, as well as patterns of functional activation revealed by standard univariate contrast methods.

We have applied this approach successfully to develop network models of single-word processing and of semantic cognition that provide a unified account of key phenomena in these domains. Yet this approach also ignores an important contribution of neural network modeling, the possibility that neural representations can be instantiated as highly distributed patterns of activation that are strongly shaped by learning.

In the second approach, I consider how statistical analysis of brain imaging data might best proceed if real neural networks have the properties predicted by neural network models. In analyses of synthetic data generated by such models, we have shown that common univariate and even popular multivariate approaches adopt implicit assumptions that prevent them from discovering essential representational structure. I will describe some new sparsity-based optimization methods that, because they begin by assuming that representations might be distributed, are better able to find such representations where they exist in the data.

When applied to real fMRI data, these methods offer a quite different picture of the nature of neuro-cognitive representations than those yielded by standard approaches. Together the work suggests that artificial neural network models can provide a useful conceptual bridge for connecting cognition and neuroscience, and that the joint application of these methods may lead to new insights about the neural bases of cognition not achievable by other means.

Novel Tools in Computational Chemistry Coding workshop (NTC3)

Sat, Apr 11, 2015

Paul Robeson Campus Center
Bergen Room

Computer hardware/software development, and computational chemistry codes originate in separate scientific communities, i.e., computer science/electrical engineering on one side, and chemistry/physics on the other side. Nowadays, the birth of novel hardware solutions, such as GPUs and coprocessors (e.g., Xeon Phi), demands a shift in the common coding paradigms in chemistry.

The NTC3 workshop is designed to promote cross-disciplinary interactions between chemistry and computer science and to set the stage for new collaborations.

The specific aim of NTC3 is to bring together chemistry and computer science experts with the goal of bringing both disciplines up-to-speed with the latest developments and main difficulties of both disciplines.

More Info :Michele Pavanello
m.pavanello@rutgers.edu
973-353-3468

Decision Models of Medical Signal and Imaging Data to Improve Medical Diagnoses

Thu, Jan 15, 2015

Aidekman Research Center (CMBN)
Aidekman seminar room

The overarching goal of our research is to develop new data analytics techniques based on applied optimization and machine learning. The main driving application of our techniques is to assist physicians in recognizing abnormality patterns (and/or patterns of interest) in medical signal and imaging data.

The main focus of our work is on feature selection, which has become an emerging problem in machine learning and optimization. Searching for the optimal set of features in decision models is computationally challenging, and it also needs to avoid model overfitting. While the main objective of most decision models is to provide an accurate decision or prediction outcome, physical/physiological interpretation of such models are extremely important in medical domain.

Our group has developed a host of feature selection techniques that can improve the accuracy and interpretability of our medical decision models. In this talk, I will discuss a few real life medical applications of our techniques, which span from prediction of neural response to visual stimuli from functional magnetic resonance imaging (fMRI), diagnostic classification of attention-deficit/hyperactivity disorder from structural (MRI), and treatment planning of lung cancer using PET/CT. If time permits, I will give an overview of other research projects undertaken in our group.

FOR MORE INFO:
Dr. Nabil R. Adam
adam@adam.rutgers.edu

Probabilistic modeling for data science and science: Nonparametric Bayes and beyond

Wed, Jan 14, 2015

Aidekman Research Center (CMBN)
Aidekman seminar room

High-dimensional datasets containing data of multiple data types have become commonplace across business, health, and science. In theory, these rich data sets support fine-grained inferences; however, the data analysis problem becomes more difficult as the data become more varied, noisy, and sparsely observed.

Moreover, people with the skills to navigate the minefield of restrictive assumptions and special-purpose analyses are scarce and expensive to train. I will describe CrossCat and BayesDB, two projects that aim to alleviate these problems. CrossCat uses nonparametric Bayesian methods from machine learning to offer generic inference for tabular data. BayesDB offers a simple language for asking questions of tabular data.

I will conclude by 1) highlighting connections to my broader research program, 2) applications of nonparametric Bayesian approaches in a variety of fields including neuroscience and geography, 3) the increasing importance of probabilistic modeling, and 4) the importance of data science skills in business, health, and science.

SPEAKERS:
Patrick Shafto

FOR MORE INFO:
Dr. Nabil R. Adam
adam@adam.rutgers.edu

Computation and Critical Dynamics in the Brain: Normal Function and Pathology

Mon, Jan 12, 2015

Aidekman Research Center (CMBN)
Aidekman seminar room

To a computer scientist, the brain is a very strange device: build of odd materials with narrow safety margins, possessed of various features our own computers are only gradually approaching, and exhibiting remarkable performance. After a review of the brain as an information processing device (its materials, speed, and performance) we proceed to discuss two hypothesized operating principles.

One, the optimality hypothesis, dominates our understanding of neuronal information processing: it is widely appreciated that the brain typically operates near theoretical limits, not only in task performance but also in adaptivity and hardware utilization.

We proceed to discuss challenges to the optimality hypothesis, and show that some troublesome phenomena can be accounted for by a second principle: the criticality hypothesis, which posits that that the brain operates in a dangerous near-critical dynamic regime; that it devotes substantial resources to avoiding super-criticality; and that various pathologies result from the criticality boundary being crossed. Phenomena discussed include epilepsy, sleep, Parkinson’s disease, and non-psychotic hallucinations (such as tinnitus, Charles Bonnet syndrome, and phantom limb) which follow diminution of sensory input.

SPEAKERS:
Barak Pearlmutter

FOR MORE INFO:
Dr. Nabil R. Adam
adam@adam.rutgers.edu