David Balduzzi and Max Planck Institute
Location: Tübingen, Germany
Abstract: This talk presents an information-geometric approach to analyzing how neurons "see the world". Neurons are modeled as probabilistic input/output devices that categorize ("measure") inputs according to the outputs they assign to them. I consider two main questions. First: How sharply do neurons categorize their inputs? This is quantified via effective information. Second: How are categorizations by neuronal assemblies composed out of sub-categorizations? The indecomposability of categorizations is quantified via integrated information, which has an interesting geometric interpretation.
Finally, time permitting, I will sketch two very different applications of these ideas. First, it turns out that effective information relates to measures of capacity (arising in statistical learning theory) that bound the expected future performance of classifiers. Second, experimental results using transcranial magnetic stimulation suggest that integrated information is higher during wakefulness than during sleep or under anesthesia. Combining these results with a simple thought experiment, I will argue that integrated information is necessary for cognitive function.
Eric Deeds University of Kansas
Abstract: Large multicomponent protein complexes, such as the ribosome and proteasome, are crucial for cellular function. Our work focuses on building computational models of the assembly of these structures. Rings represent an important class of structural motifs; they can display remarkable thermodynamic stability that causes the overall assembly reaction to approach completion. Our model of ring assembly indicates that the dynamics of this process can display complex behaviors. We have found that rings can optimize assembly according to a wide range of criteria by exhibiting at least one protein interaction that is significantly weaker than the others in the ring. Analysis of the experimentally available structures of heteromeric 3-membered rings indicates that most have evolved such a weak bond, as we would predict. We have also examined the process of complex formation in the context of large protein-protein interaction and signaling networks. These networks are combinatorially complex, in the sense that they can generate astronomical numbers of possible molecular species. We employed a recently developed rule- and agent-based modeling technique to simulate the dynamics of two large networks. Our results indicate that the combinatorial complexity of this network engenders "drift" in the space of molecular possibilities. To produce large complexes that assemble reliably into well-defined, stable structures, cells have had to evolve mechanisms that constrain and eliminate this drift.
Sam Landry Tulane University School of Medicine
Abstract: The human immune system repels infectious pathogens by attacking certain immunodominant molecular structures in the pathogens. However, pathogens like the human immunodeficiency virus (HIV) may have used molecular structures to misdirect the helper T cells. This research would reveal the rules governing how the immune system chooses targets for helper-T-cell immunity and how it decides whether to use antibodies or cytotoxic T cells in the fight against infection.
Sponsored By: 2012 D. W. Mitchell Lecture Series and the Provost's Faculty Seminars in Interdisciplinary Research.
Aron Culotta Southeastern Louisiana University
Abstract: The proliferation of social media (Twitter, Facebook, blogs, etc.) has created an unprecedented, continuous stream of messages containing the thoughts of millions of people. The nascent field of Social Media Analysis (SMA) combines natural language processing, data mining, machine learning, and statistics to explore what we can infer from the behavior of social media users. Recent research suggests that such analysis can provide insights into public health, finance, politics, social unrest, and natural disasters. In this sense, SMA can be understood as an alternative to slower and more costly data collection methods, such as surveys and opinion polls.
In this talk I will first give an overview of SMA methodology, then present results from three recent applications: (1) estimating national influenza rates, (2) estimating alcohol consumption volume, (3) assessing personal risk perception prior to an impeding natural disaster. These results suggest that relatively simple methods can extract socially valuable insights from this rich source of data. I will conclude with a discussion of open problems and discuss how more sophisticated machine learning algorithms (graphical models, semi-supervised learning) may expand the capabilities of this emerging field of study.
Ken Ford Florida Institute for Human and Machine Cognition
Abstract: The emerging concept of human-centered computing represents a significant shift in thinking about intelligent machines and, indeed, about information technology in general. My talk will provide a survey of selected research activities at the Institute for Human & Machine Cognition (IHMC) developed under this framework. Human Centered Computing research requires a broader interdisciplinary range than is typically found in one organization, and IHMC staff includes computer scientists, cognitive psychologists, neuroscientists, physicians, philosophers, engineers and social scientists of various stripes, as well as some people who resist all attempts to classify them.
Carl Baribault Tulane Cancer Center
Abstract: Attempts to find novel genes using pure homology-based methods are expected to remain insufficient largely due to the roughly half of all genes in eukaryotic (intron-bearing) genomes being specific to the given organism, such as in the nematode C. Elegans, a model eukaryotic genome. Also, the simple, 1st-order Hidden Markov Model is insufficient to leverage the typically extended range of information (as measurable via Shannon entropy) in the vicinity of the type-specific, coding-noncoding boundaries of a eukaryotic genome. My talk will cover some of the intricacies involved in the development of this (meta-state) HOHMM for the prediction of eukaryotic, protein-coding genes, some issues for the standards of comparison among various prediction tools, and some comments on future extension of the HOHMM to include support for transcription-promoting motifs and perhaps non-coding genes as well.
Carola Wenk UTSA
Abstract: Geometric shapes are at the core of a wide range of application areas. In this talk we will discuss how approaches from computational geometry can be used to solve shape matching problems arising in a variety of applications including biomedical areas and intelligent transportation systems. In particular, we will discuss point pattern matching algorithms for the comparison of 2D electrophoresis gels, as well as algorithms to compare and process trajectories for improved navigation systems and for live cell imaging.
Brent Venable University of Padova, Italy
Abstract: As preferences are fundamental for the analysis of human choice behavior, they are becoming of increasing importance for computational fields such as artificial intelligence (AI). Their embedding in intelligent systems calls for both expressive and, at the same time, compact representation models as well as for efficient reasoning machinery.In this talk we will start by giving a brief overview of soft constraints and CP-nets: two of the most successful AI compact preference frameworks currently used to represent the preferences of a single agent. We will discuss how, for example, uncertainty about preferences can be efficiently dealt with within such frameworks. We will also show how such models can be embedded in multi-agent settings, such as decision making via voting, by equipping them with appropriate reasoning tools and adapting voting protocols to compact preference structures. Finally, we will conclude by highlighting some promising and exciting future research directions in this field.
School of Science and Engineering, 201 Lindy Boggs Center, New Orleans, LA 70118 504-865-5764 email@example.com