Wesleyan Home → Mathematics and Computer Science → Past Events → Computer Science Seminar

Seminars and Colloquia

Computer Science Seminar

Tuesday, April 09, 2013

04:30 pm - 05:30 pm

Computer Science Seminar, Kegan Samuel: "How do computers 'see'?"

Abstract: The human visual system can perceive and understand the world around with apparently very little difficulty. In the computer vision field, researchers attempt process and understand digital representation of images in a similar manner using various computer algorithms or mathematical models. While much has been accomplished in the field, there is still a significant amount of work to be done before a system is developed which would be able to interpret images at a level close to that of humans. In addition to a brief overview of the problems encountered in the field of computer vision, I will also speak on work dealing with the use of Markov Random Field(MRF) models applied to the tasks of denoising and image segmentation.

ESC 638

Tuesday, May 01, 2012

04:15 pm - 05:30 pm

Computer Science Seminar, Stefan Sundseth '13: The Pi-Calculus for Mobile Systems

Abstract: In today's world computation revolves around interaction between multiple systems, yet the most well-known models of these computational systems neglect their ability to communicate with external processes. The Pi-Calculus was created as a means to rigorously describe a model whose fundamental action is to send and receive messages across an interface, which means that multiple processes can be synchronized by concurrent send/receive directives. Though simple, the Pi-Calculus is actually quite powerful as concepts such as algorithms, object-oriented programming, and the lambda-calculus fall within its scope.

ESC 339

Tuesday, April 24, 2012

04:15 pm - 05:30 pm

Computer Science Senior Thesis Talks: Jennifer Payking, Jeffrey Ruberg and Micah Wylde

Abstracts: Jennifer Paykin: Automated Cost Analysis of a Higher-Order Language in Coq The cost analysis of a program is a function from the size of its input to the number of steps the program takes to run. In this work we address two difficulties associated with constructing cost analyses. First, what is the cost of a higher-order program? If an expression takes a function as input, what is the size of that input? If an expression evaluates to a function, what is its cost? Second, how can we automate the cost analysis of arbitrary programs? In this work we develop a static cost analysis system for a functional target language with structural list recursion. A translation function maps target expressions e to complexities encoding upper bounds on both the cost of evaluating e and some measure of e's size. We implement the translation in the proof assistant Coq, formally verifying the soundness of the translation scheme. The resulting system generates certified upper bounds on the complexity of target language terms. Jeffrey Ruberg: Charming a Python with Static Type Checking Pyty is a static typechecker for the Python programming language. Pyty imposes a static type system on users, for which they provide type declarations via comments. The type system developed is modeled on the Hindley-Milner type inference algorithm for the lambda calculus and its application to ML. Much of the implementation challenge is in applying this type system to a language not designed with typechecking in mind. The type system describes a system of deriving type assignments, whereby we can prove that a Python term can be assigned a specific type; we do this by searching for a valid tree of type assignment rules. Pyty also employs type inference for a restricted subset of the Python language in order to facilitate the typechecking algorithm. Micah Wylde: Safe Motion Planning for Autonomous Driving Self-driving cars have the potential to revolutionize transportation by making it cheaper, safer and more efficient. In this work we describe a novel motion planning system for translating high-level navigation goals into low-level actions for controlling a car. Specifically, the motion planning system is responsible for choosing at each time step an appropriate velocity and steering angle, which could then be implemented by the driving hardware or simulation. Our planner is able to compute a safe and efficient trajectory in a dynamic environment while staying within its lane and avoiding obstacles.

ESC 339

Tuesday, February 28, 2012

04:15 pm - 05:15 pm

Computer Science Seminar, Matt Eaton '04, Computer Science and Artificial Intetelligence Lab, MIT: "Chromatin and epigenetic regulation in DNA replication and Alzheimer's disease"

Abstract: In the nucleus of every eukaryotic organism, DNA is compacted and organized by wrapping around histone proteins in a structure called chromatin. Chromatin is a dynamic structure, and encoded in the histone tails are chemical modifications that provide clues to the underlying regulation of DNA-templated processes such as RNA transcription and DNA replication. This "histone code" can be measured genome-wide with high-throughput assays such as ChIP-sequencing. A further layer of information is encoded in chemical modifications to the DNA itself; for example, cytosine residues in the genomes of higher eukaryotes can be methylated by nuclear proteins in heritable fashion, and this methylation can alter gene expression. Taken together, these layers of information encoded on top of the DNA sequence make up a large part of the field of epigenetics: the study of heritable, non-sequence, cis-acting changes to nuclear DNA and chromatin. In this talk, I will give three examples of how machine learning and computer science techniques were used to help elucidate the epigenetic landscape in both model organisms and human disease. The first will examine the role of nucleosome positioning around origins of DNA replication in the genome of the budding yeast, Saccharomyces cerevisiae. The second example examines the role of the histone code in specifying the replication program of the fruit fly Drosophila melanogaster. Finally, I will discuss ongoing work in examining the role of DNA methylation in the cognitive decline of Alzheimer's disease patients.

ESC 113

Thursday, February 23, 2012

12:00 pm - 01:00 pm

(Computer Science Seminar) Matt Ginsberg, CEO On Time Systems, Inc: "Dr. Fill: Crosswords and An Implemented Solver for Singly Weighted CSPs"

Abstract: We describe and demonstrate Dr.Fill, a program that solves American-style crossword puzzles. From a technical perspective, Dr.Fill works by converting crosswords to weighted constraint satisfaction problems (CSPs), and then using a variety of novel techniques to find a solution. These techniques include generally applicable heuristics for variable and value selection, a variant of limited discrepancy search, and postprocessing and partitioning ideas. Branch and bound is not used, as it was incompatible with postprocessing and was determined experimentally to be of little practical value. Dr.Filll's performance on crosswords from the American Crossword Puzzle Tournament suggests that it ranks among the top fifty or so crossword solvers in the world.

ESC 339

Wednesday, November 30, 2011

04:15 pm - 06:00 pm

Dan Dougherty, Wes and Worcester Polytechnic Institute: Symbolic protocol analysis for implicitly authenticated Diffie-Hellman

Abstract: Joint work with Joshua Guttman, WPI. We develop new techniques for rigorous symbolic proofs of security properties for implicitly authenticated Diffie-Hellman protocols. These protocols, e.g.~{MQV} and the Unified Model, which construct a shared secret using a minimum of communication, have been challenging both for symbolic and computational analyses. Their essential use of rich algebraic structure has been an obstacle. We prove implicit authentication and secrecy results for several protocols in this group. The same methods yield proofs of forward secrecy, for protocols where it holds, and indicate why it does not otherwise. Resistance to compromised key attacks and impersonation attacks may be resolved similarly. Our crucial notion is the \emph{indicator}, a vector of integers that summarizes how many times each secret exponent appears in a message. We prove that the adversary can never construct a message with a new indicator in our adversary model. Although we work with a message algebra generated by the normal forms for an equational theory $AG^$, we also give a model-theoretic proof that $AG^$ faithfully reflects all equations that are true uniformly as the prime $q$ varies. Hence, our security theorems ensure that the adversary has no strategy that is expressible algebraically and works uniformly---independent of the choice of $q$---for producing a counterexample. Computational soundness awaits further investigation.

ESC 638

Tuesday, November 01, 2011

04:15 pm - 05:30 pm

Comp Seminar: Summarizing the News

Speaker: Dan Gillick, '03, UC Berkeley and Google Abstract: The goal of automatic summarization is to generate a short summary of a document, or a set of related documents. This is hard for people, let alone computers. I'll describe how these systems work, how they are evaluated, and how I improved them. Hopefully, there will be a demo.

ESC 339

Tuesday, October 18, 2011

04:15 pm - 05:30 pm

Computer Science Seminar: Abstract notions of computation and monads Part 2

Speaker: Emilio Jesus Gallego Arias, UPM Madrid and Wesleyan University Abstract: Every day, thousands of programmers alien to category theory use monads as a fundamental tool to their work. In fact, in popular functional languages like Haskell you cannot write a program without using a monad. At the time monads were first studied in the context of category theory, computer science didn't have a need for them. 20 years after that, a new kind of functional programming languages was born, advocating the usage of Scott's computable functions as the main primitive for program construction. Semantics based on mathematical functions have numerous advantages, but essential needs such as input/output, global state, error handling, etc... couldn't be solved in an elegant way due to the lack of side effects and order or evaluation issues. We will start with categorical semantics for functional programming and monads understood as a "notion of computation". This allows otherwise impure or ugly constructions. We will discuss the evolution of monads, from the first theoretical papers to real world use and implementation. We will focus on how the original notion led to an "abstract notion of computation" or "abstract computation types", akin to the popular "abstract data types" paradigm. Last, we will present joint work with Prof. Lipton about incorporating the notion of "abstract computation" to logic programming.

ESC 339

Thursday, October 06, 2011

11:30 am - 02:00 pm

WANTED! CS Majors to Study Abroad

Speaker: Dr. Andras Recski, Director of Academic Programs, AIT BUDAPEST An upcoming talk will be presented on campus to encourage students to consider a great new study abroad program, Aquincum Institute of Technology BUDAPEST, for students interested in computing, design, computational biology, and IT entrepreneurship. Andras Recski, Director of Academic Programs at AIT, will be giving a presentation about the program on THURSDAY, OCTOBER 6, 2011 including a NOON LUNCH. Pizza and treats from Hungary will be served! About AIT: The AIT program has a first-rate faculty including professors such as Erno Rubik (inventor of the Rubik's Cube and recent recipient of the U.S. Outstanding Contributions to Science Education Award), an innovative curriculum including courses such as "Computer Vision for Digital Film Post-production" taught by faculty affiliates from Colorfront Studios (recent recipients of an Academy Award for technical contributions), and a guest lecture series that brings prominent speakers to campus. All classes are conducted in English at AIT's state-of-the-art campus on the lovely banks of the Danube River. Students live in vibrant neighborhoods of Budapest and have ample opportunities to interact with Hungarian students and explore Hungary and the region. AIT is small and friendly, with typical class sizes of 5-15 students. Recent U.S. AIT students have come from Franklin W. Olin College of Engineering, Harvey Mudd College, Northeastern University, Pomona College, Princeton University, RPI, Skidmore, Smith, Swarthmore and Williams Colleges. The program also includes a small number of Hungarian students. AIT Alumni site: http://www.ait-budapest.com/ait-alumni The AIT website and APPLICATION materials are available on-line at: http://www.ait-budapest.com Applications for the Spring 2012 term are due by November 15, 2011. (Although, completed submissions are currently being accepted and reviewed. Notice is then given within two weeks time.) Contacts: Prof. Ran Libeskind-Hadas (hadas@cs.hmc.edu) and Prof. Michael Orrison (orrison@math.hmc.edu) at Harvey Mudd College are serving as the North American Co-Directors for AIT, and are eager to answer any questions that you might have.

ESC 184 (Woodhead Lounge)

Tuesday, October 04, 2011

04:15 pm - 05:30 pm

Abstract notions of computation and monads

ESC 339

Tuesday, April 26, 2011

04:15 pm - 05:30 pm

A dynamical systems-based model for indoor robot navigation

Speaker: Juan Pablo Mendoza, Wesleyan, M.A. Thesis Defense Abstract: Common indoor environments are particularly challenging scenarios for autonomous robot navigation: robots may need to navigate in relatively tight and complicated paths while being responsive to unpredictable changes in the environment, such as those created by humans walking around. In my talk, I will describe a new, dynamical systems-based model for autonomous indoor navigation. In the dynamical systems approach to behavior modeling, safe navigation around obstacles and toward a goal is achieved by letting carefully designed differential equations control the evolution in time of the robot's heading direction as it navigates through its environment. Unlike previous dynamical systems-based navigation models, this one focuses on addressing the problems of navigation in indoor environments. Simulations and real robot results support the effectiveness of the navigation model presented. Only very basic knowledge of differential equations will be assumed.

ESC 339

Monday, April 25, 2011

04:15 pm - 06:00 pm

Dynamical BDI-Based Intention Recognition for Virtual Agents

Speaker: Ted Nichols, '10, M.A. Thesis Defense Abstract: Virtual agents might need to recognize the intentions of other agents in order to adapt to a dynamic environment. In this talk, I will discuss ways to incorporate intention recognition into the cognitive system of a hybrid dynamical cognitive agent (HDCA). HDCAs have cognitive systems that consist of beliefs, desires, and intentions that evolve dynamically over time. While intention recognition methods for Belief-Desire-Intention (BDI) architectures have been implemented in the literature, none have been based on dynamical BDI systems. Our work was motivated by the intention recognition tasks that might face an informal college tour guide; adding intention recognition beliefs to a guide's cognitive system allows the guide to respond dynamically to the intentions of a tour taker.

ESC 339

Wednesday, April 20, 2011

04:15 pm - 05:30 pm

Simulating circuit creation in Tor

Speaker: William Boyd, Wesleyan, Senior Thesis Presentation Abstract: All computer security systems are subject to attacks. When attacks are proposed, they must be tested to determine practicability in real-world usage (and thus whether defenses are needed or possible). It is not always possible, practical or ethical to test attacks on the real systems they target; thus, attacks are often tested in simulation. We have found in our research on the anonymity network Tor that many simulations make unreasonable or unjustified assumptions; we thus attempt to produce an accurate, justified simulation on which to easily test attacks. I'll be describing and justifying the methods we use to simulate circuit creation, a particular attackable aspect of the behavior Tor. We directly observe the behavior of the extant network, and attempt to construct a set of hidden Markov models which accurately simulate this behavior. Particular challenges we've faced include determining a suitable description of network behavior, simplifying our observations to make HMM generation computationally feasible, and objectively evaluating the accuracy of our simulation.

ESC 339

Tuesday, April 19, 2011

04:15 pm - 05:30 pm

Challenges in Internet Investigations

Speaker: Marc Liberatore, UMass Amherst Abstract: Law enforcement, computer security professionals and other investigators work to discover, document, and potentially prosecute crimes facilitated by the Internet. This work ranges widely, encompassing tasks such as observing new malware, documenting network intrusions, or trafficking in digital contraband. Recently, my colleagues and I have developed a suite of tools to ease investigation by law enforcement of contraband trafficking on peer-to-peer filesharing networks. These tools generate a plethora of leads for investigators, and it is now clear that determining the dangerousness of suspects is of key importance. This problem, of modeling suspects' behavior on the basis of many observations over time, is an example of a question of computational social science. In this talk, I will describe the basis of our tools and measurements, touching upon the legal and practical aspects of these investigations. I will also present results of our measurements over the last 15 months, and discuss the challenges ahead.

ESC 339

Thursday, April 07, 2011

04:15 pm - 05:15 pm

Decisions, Decisions

Speaker: Eli Fox-Epstein, '10 Abstract: In this talk, I'll describe the forbidden pairs problem. When there are constraints on which vertices or edges one can employ, finding a subgraph with a property can become much more difficult. We generalize a proof technique to describe the complexity of infinitely many problems, instead of individual problems.

ESC 339

Tuesday, March 29, 2011

04:15 pm - 05:30 pm

TBA

Speaker: Marc Liberatore, UMass Amherst Abstract: TBA

ESC 339

Tuesday, March 22, 2011

04:15 pm - 05:30 pm

It's Not "Junk": Using Genetic Algorithms to Search for Functional Regions in Noncoding DNA

Speaker: Clare Bates Congdon, '85, University of Southern Maine Abstract: Over 95 percent of our DNA does not directly code for genes. Formerly called "junk" DNA, it is now understood that some of these noncoding regions are functional... but determining which regions are functional is a complex problem. Our hypothesis is that regions of noncoding DNA that have been conserved across evolutionary time are good candidates as functional elements. As computer scientists, our role is to identify patterns (aka motifs) that appear to be conserved through evolution. To find these motifs, we use a search approach called genetic algorithms, in which one "evolves" the solution to a problem. In this talk, I'll describe GAMI, a genetic algorithms approach to motif inference. The talk will describe the motif inference problem, the genetic algorithms approach in general, and the specific GAMI system design, as well as some of the datasets that we have studied using this approach, and our recent results.

ESC 339

Thursday, February 24, 2011

12:00 pm - 01:00 pm

A review and update on Tor

Speaker: Andrew Lewman, The Tor Project Abstract: A review and update on Tor, how open source solutions work well worldwide, where we're headed, and where we need help from developers like you. Why Tor? Tor is a tool to protect your online privacy and anonymity. We rely on thousands of volunteers to run our network, review our code, and help enhance the experience for all. Lunch will be served.

ESC 184 (Woodhead Lounge)

Tuesday, February 15, 2011

04:15 pm - 05:30 pm

Combinatorics of RNA Secondary Structures and Related Problems

Speaker: Danny Krizanc, Wesleyan Abstract: We derive expressions for the asymptotic number of RNA secondary structures satisfying a variety of constraints. We also discuss the asymptotics of the expected number of base pairs in randomly chosen structures, again, under different constraints. Finally, we relate these results to problems arising in the analysis of algorithms.

ESC 339

Tuesday, November 16, 2010

04:15 pm - 05:30 pm

Comparison of Ecotype Demarcation Algorithms

Speaker: Carlos Francisco, Wesleyan '11 Abstract: For a long time, microbiology has concerned itself with its own version of the 'species problem' - that there are multiple proposed definitions of what exactly species are and that biologists have yet to agree upon which definition calls for the most sensible criteria. For microbiologists, the problem is exacerbated by bacteria existing in very large populations and accumulating mutations at a much higher rate, allowing them to diverge much more quickly. A named species of bacteria might be extremely diverse genotypically and phenotypically, one example being /Escherichia coli. / Within a named species, there might be multiple clusters of bacteria that are adapted to a particular ecological niche. We call these groups 'ecotypes.' It would be of great benefit to be able to demarcate the ecotypes in a set of bacteria, and with the advent of gene and full genome sequencing, we are now able to do this algorithmically. In this talk, I will present four distinct algorithms that tackle the problem of figuring out which bacterial strains are in which ecotypes. To compare their qualities, we first used an algorithm that generates simulated nucleotide strains and the 'correct' ecotypes for those strains. We then ran each ecotype demarcation algorithm on those strains and compare them to the 'real' ecotypes using a metric called variation of information. I will talk about the algorithms involved and how they work, the accuracy of the results each algorithm produced, the running times of the algorithms, and I will also provide an example of how each algorithm runs on a set of real, sequenced bacterial data.

ESC 339

Friday, October 29, 2010

12:00 pm - 01:15 pm

Algorithmic Detection of Computer Generated Text and Measures of Paper Citations

Speaker: M.S. Krishnamoorthy, RPI Abstract: The talk consists of two parts. The first part discusses Detection of Computer Generated Text. Computer generated academic papers have been used to expose a lack of thorough human review at several computer science conferences. We assess the problem of classifying such documents. After identifying and evaluating several quantifiable features of academic papers, we apply methods from machine learning to build a binary classifier. In tests with two hundred papers, the resulting classifier correctly labeled papers either as human written or as computer generated with no false classifications of computer generated papers as human and a 2% false classification rate for human papers as computer generated. The second part discusses the measures of citations of Scientific Papaers. A new way of computing the value of citations is introduced and compared with the PageRank algorithms for author ranking. In our proposed approach, the value of each paper is expressed in CENTs (scientific currency tokens). The paper.s value is then divided by the number of citations of that paper yielding the value for each citation. As citations are the acknowledgment of the value of the work of authors other than oneself (such citation is equivalent to full acknowledgment of a publication that has been useful), then self citations count as zero acknowledgment. Circular citations, as a generalized type of self citations, will be considered at reduced acknowledgment value smaller than a full one. Finally, we propose a redefinition of the h-index, as the largest integer such that the h-th paper (on the list of papers sorted by their CENTs values) is worth more than the h CENTs. The new index, termed i-index or i2 in short, appears to be a more precise measure of the impact of authors. papers and their productivity than the h-index.

ESC 339

Monday, October 11, 2010

12:00 pm - 01:00 pm

Dynamical Intelligence: Navigation and Intention

Speaker: Eric Aaron, Wesleyan University Abstract: Intelligent computational \emph{embodied agents}, whether virtual (characters in animated worlds) or physical (mobile robots), move through their worlds, performing sequences of tasks to achieve their goals. To succeed in complicated or unpredictable environments without human intervention, these agents must be autonomously responsive and adaptive in their worlds--for instance, they need to avoid collisions with obstacles as they move, and they may need to adjust the sequences of tasks they intend to perform to reflect new priorities. In this talk, I present dynamically responsive approaches to collision-free navigation and intention-guided task sequencing for embodied agents, and I describe \emph{hybrid dynamical cognitive agents} (HDCAs), which smoothly integrate these navigation and task-sequencing levels of intelligence. By enhancing low-level, dynamically sensitive intelligence without minimizing high-level, logic-based intelligence, the underlying HDCA framework can extend some capabilities of embodied agents, and it may have implications for related learning-based, developmental, and dynamicist approaches to behavior modeling.

ESC 339

Tuesday, September 28, 2010

04:15 pm - 05:30 pm

Detecting and Defending Against Attacks on Tor

Speaker: Sam DeFabbia-Kane Abstract: Tor is a low-latency anonymity network used worldwide by governments, dissidents, journalists, and other people who can benefit from anonymity online. Tor works by forwarding its users' connections through a series of volunteer-run Tor routers, so that a single centralized authority cannot compromise Tor's anonymity. However, this design does leave Tor open to several kinds of attacks by people running or observing Tor routers. This talk focuses on one of those attacks: an end-to-end correlation attack whereby an attacker can identify a user by comparing timing data taken from data streams at the beginning and end of the Tor circuit. The talk will cover some variations on the attack and our efforts at detecting attackers implementing one variation in particular: a denial of service attack proposed by Borisov et al., as well as our future plans for investigating the effectiveness of end-to-end correlation on the Tor network itself.

ESC 339

Thursday, September 16, 2010

04:15 pm - 05:30 pm

Software Engineering Using RATionale

Speaker: Janet Burge, Miami University, Ohio Abstract: Abstract: Many decisions have to be made when developing a software system and a successful outcome depends on how well thought out these decisions were. The decisions made, and alternatives considered, form the rationale for the system. The rationale goes beyond standard documentation by describing the developers' intent and all alternatives considered rather than only those selected. While the potential usefulness of this information is seldom questioned, the rationale is rarely captured in practice. There is a pervasive belief that developers will not be willing to take the time and effort to perform what might be perceived as extra documentation. In order to encourage rationale capture there needs to be some incentive to do so. This talk describes the Software Engineering Using RATionale system (SEURAT). SEURAT is integrated with the Eclipse Interactive Development Environment and inferences over the rationale to evaluate decision alternatives and perform impact assessment when requirements, development criteria, and assumptions change. In addition to development environment integration, SEURAT also supports importing rationale extracted from external sources, such as Word documents, to help further reduce the effort of rationale capture. SEURAT also interfaces with the XFeature feature modeling tool so that rationale can be used to help customers use the rationale to guide product feature selection. Biography: Janet Burge is an Assistant Professor in the Miami University Computer Science and Software Engineering department. She received her Ph.D. in Computer Science from Worcester Polytechnic Institute (2005) and performed her undergraduate work at Michigan Technological University (1984). Her research interests include design rationale, software engineering, AI in design, and knowledge elicitation. She is a co-author (with Jack Carroll, Ray McCall,and Ivan Mistrik) of the book Rationale-Based Software Engineering. Dr. Burge is a recipient of a NSF CAREER Award for her project Rationale Capture for High-Assurance Systems. She has been at Miami University since 2005. Prior to that point, she worked for more than 20 years in industry as a software engineer and research scientist.

ESC 638