Integrated Calendar

Several state-of-the-art computer vision systems for face detection, e.g., Viola-Jones [1], rely on region-based features that compute contrast by adding and subtracting average image intensity within different regions of the face. This is a powerful strategy due to the invariance of these features across changes in illumination (as proposed by Sinha [2]). The computational mechanisms underlying face detection in biological systems, however, remain unclear. We set to investigate the role of region-based features in the macaque middle face patch, an area that consists of face-selective neurons. We found that individual neurons were tuned to subsets of contrast relationships between pairs of face regions. The sign of tuning for these relationships was strikingly consistent across the population (for example, almost all neurons preferred a lower average intensity in the eye region relative to the nose region). Furthermore, the pairs and polarity of tuning were fully consistent with Sinha’s proposed ratio-template model of face detection [2]. Non-face images from the CBCL dataset that contained correct contrast polarities in pre-defined regions (facial parts) did not elicit increased firing in face-selective neurons, suggesting that the neurons are not only computing averaged intensity according to a fixed template, but are also sensitive to the specific shape of features within a region.

[1] Robust Real-time Object Detection, Paul Viola and Michael Jones.
Second International Workshop on Statistical and Computational Theories of Vision – Modeling, Learning, Computing, and Sampling.
Vancouver, Canada, July, 2001.
[2] Qualitative Representations for Recognition, Pawan Sinha. Proceedings of the Second International Workshop on Biologically
Motivated Computer Vision, Tubingen, November, 2002.

Many ideas from theoretical physics can be brought over to queueing theory. In this talk we shall go over recent (and not so recent) work on the borderline between queues and physics. In particular we shall compare queueing networks to reaction diffusion problems and make analogies to the “second quantized” approach, mean field and Bethe ansatz solutions.

Of course, to do this, we shall include a brief introduction to queueing theory and an even briefer introduction to second quantization. Hopefully both queueing theorists and physicists will find each others’ discipline fascinating.

I will present a holographic description of N=1 supersymmetric QCD with a large number of colors and flavors. The gauge theory has multiple Higgs and confining vacua. I will focus on the confining vacuum, and use the holographic description to calculate the spectrum of mesons and other properties. I will also describe some generalizations of the construction.

Our group is focused on the computational underpinning of genomics, developing new algorithms and machine learning techniques for studying complete genomes, understanding their regulatory constructs, and their evolutionary dynamics. We have defined evolutionary signatures in the nucleotide alignments of multiple related species, enabling the systematic discovery and characterization of diverse classes of functional elements, including protein-coding genes, RNA structures, microRNAs, developmental enhancers, regulatory motifs, and biological networks. We have also defined distinct chromatin signatures from combinations of many histone marks in genome-wide epigenomic datasets, revealing numerous classes of promoter,enhancer, transcribed, and repressed regions, each with distinct functional properties. Lastly, we have defined activity patterns of gene expression, chromatin state, motif enrichment, and positional constraints across multiple cell types to link enhancers to candidate targets, recognize cell-type specific activators and repressors, and evaluate predictive models of gene regulation. These techniques have enabled us to discover many new insights into animal gene regulation, in the context of the ENCODE, modENCODE, and Epigenome Roadmap projects, and in the comparative analysis of many Drosophila and mammalian genomes.

Autophagy is a major catabolic pathway by which eukaryotic cells degrade and recycle macromolecules and organelles. This pathway is activated under environmental stress conditions, during development and in various pathological situations, including cancer. A growing body of data suggests a role for the p53 tumor suppressor gene in regulation of autophagy. The nature of this regulation is complex, involving transcriptional activation of autophagy related genes, as well as inhibition of autophagy by cytoplasmic p53. In my talk I will describe new findings showing that p53 supports cell survival under chronic starvation conditions by downregulating the expression of LC3, a pivotal component of the autophagic machinery; this results in reduced but otherwise normal autophagy. Conversely, lack of p53 causes accumulation of excessive LC3 and formation of aberrant autophagosomes, which fail to undergo lysosomal fusion and degradation. Consequently, autophagic flux is severely impaired in p53-deficient cells, leading to increased apoptosis. Thus, p53 provides a survival function under severe starvation by maintaining sustainable autophagic flux.