Select a calendar:
Filter October Events by Event Type:
SUNMONTUEWEDTHUFRISAT
Conferences, Lectures, & Seminars
Events for October
-
CS Colloquium: Ardalan Amiri Sani (UC Irvine) - Dealing with Vulnerabilities in Device Drivers
Wed, Oct 03, 2018 @ 11:00 AM - 12:20 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Ardalan Amiri Sani, UC Irvine
Talk Title: Dealing with Vulnerabilities in Device Drivers
Series: Computer Science Colloquium
Abstract: Vulnerabilities in the device drivers of today's commodity operating systems (e.g., Android) remain a security concern due to the monolithic structure of the kernel. In this talk, we investigate three methods to mitigate this concern, with a focus on mobile devices. These approaches include a novel tool to find and fix these vulnerabilities, an efficient vetting layer to make exploits harder, and a device driver design possible for I/O devices with virtualization support.
This lecture satisfies requirements for CSCI 591: Research Colloquium. Please note, due to limited capacity, seats will be first come first serve.
Biography: Ardalan Amiri Sani is an Assistant Professor in the Computer Science department at UC Irvine. His research is at the intersection of mobile computing, security, and operating systems. His work has appeared in various top-tier conferences such as MobiSys (including a best paper award), USENIX Security, CCS, and ASPLOS. Ardalan received his Ph.D. from Rice University in 2015.
Host: Ramesh Govindan
Location: Olin Hall of Engineering (OHE) - 100D
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Mohammad Soleymani (USC-ICT) - What Do Machines Learn in Emotion Recognition from EEG Signals?
Thu, Oct 04, 2018 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Mohammed Soleymani, USC-ICT
Talk Title: What Do Machines Learn in Emotion Recognition from EEG Signals?
Series: CS Colloquium
Abstract: Machines that are able to read our emotions and cognitive states make better companions. Emotions are often sensed by their external manifestations such as facial and vocal expressions. Additionally, studies in affective neuroscience have identified a set of emotional neural activities that can be captured by eletroencephalogram (EEG) signals, including asymmetric frontal brain activity and increase in information transfer. Motivated by these findings, a growing number of studies report developing EEG-based emotion recognition systems with promising results. In this talk, I first present my work on recognizing emotions of people watching videos. I then present a follow up study in which we aimed to better understand what machine learns in such scenarios. In the follow up work, we recorded a dataset which includes spontaneous emotions and posed expressions. Our analysis on the data collected in the follow up study demonstrates that the performance of existing EEG-based emotion recognition methods significantly decreases when evaluated across different corpora. We also found that models trained on spontaneous emotions perform well on recognizing mimicked expressions. Our results provide evidence that stimuli-related sensory information and facial electromyogram activities are the main components learned by machine learning models for emotion recognition using EEG signals.
This lecture satisfies requirements for CSCI 591: Research Colloquium. Please note, due to limited capacity, seats will be first come first serve.
Biography: Mohammed Soleymani is a research scientist with the USC Institute of Creative Technologies. He received his PhD in computer science from the University of Geneva in 2011. From 2012 to 2014, he was a Marie Curie fellow at Imperial College London. Prior to joining ICT, he was a research scientist at the Swiss Center for Affective Sciences, University of Geneva. His main line of research involves developing automatic emotion recognition and behavior understanding methods using physiological signals and facial expressions. He is also interested in understanding subjective attributes in multimedia content, e.g, predicting whether an image is interesting from its pixels or automatic recognition of music mood from acoustic content. He is a recipient of the Swiss National Science Foundation Ambizione grant and the EU Marie Curie fellowship. He has served on multiple conference organization committees and editorial roles, most notably as associate editor for the IEEE Transactions on Affective Computing and technical program chair for ACM ICMI 2018 and ACII 2017. He is one of the founding organizers of the MediaEval multimedia retrieval benchmarking campaign and the president elect for the Association for Advancement of Affective Computing (AAAC).
Host: David Traum
Location: Ronald Tutor Hall of Engineering (RTH) - 115
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CS Tech Talk: Parisa Mansourifard (Facebook) - Infrastructure Data Science Team at Facebook
Thu, Oct 04, 2018 @ 03:30 PM - 04:50 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Parisa Mansourifard, Facebook
Talk Title: Infrastructure Data Science Team at Facebook
Series: Computer Science Colloquium
Abstract: In this talk, I will present what my team does at Facebook and what problems we aim to solve. Infrastructure Data Science partners with engineering teams to develop data-driven solutions for significant infrastructure challenges such as app and site performance, systems efficiency and reliability, resource allocation and long-term capacity forecasts. Infra Data Scientists use a range of tools, from A/B testing to machine learning, to help Facebook make decisions about operations and system design. The team contributes to all parts of a project's lifecycle, including scoping, data discovery, research, methodological design, code implementation, and reporting and interpreting final results. The teams' work varies, in line with the complex and diverse challenges of maintaining one of the largest and most advanced enterprise infrastructures in the world. We look for candidates with a wide range of backgrounds to join our team and help with this work.
This lecture satisfies requirements for CSCI 591: Research Colloquium.
Biography: Parisa Mansourifard is currently a data scientist at Infrastructure data science team at Facebook. Before joining Facebook, she was a data scientist at SupplyFrame Inc. and a part-time lecturer at CS department of University of Southern California teaching machine learning. She received the B.S. and M.S. in electrical engineering from Sharif university of technology, Tehran, Iran, in 2008 and 2010 respectively. She also got a M.S. in computer science and Ph.D. in electrical engineering from University of Southern California, Los Angeles, CA, USA, in 2015 and 2017, respectively. During her Ph.D. she held Viterbi Dean fellowships in 2011-2014 and AAUW dissertation completion fellowship in 2015-2016. She also got a best paper award for the operations research track at EU IEOM conference in Paris 2018.
Host: Computer Science Department
Location: Henry Salvatori Computer Science Center (SAL) - 101
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Xinyu Xing (Pennsylvania State University) - Tracking down Software Vulnerabilities from Unexpected Crashes
Tue, Oct 09, 2018 @ 03:40 PM - 04:50 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Xinyu Xing, Pennsylvania State University
Talk Title: Tracking down Software Vulnerabilities from Unexpected Crashes
Series: Computer Science Colloquium
Abstract: Despite the best efforts of developers, software inevitably contains flaws that may be leveraged as security vulnerabilities. Modern operating systems integrate various security mechanisms to prevent software faults from being exploited. To bypass these defenses and hijack program execution, an attacker therefore needs to constantly mutate an exploit and make many attempts. While in their attempts, the exploit triggers a security vulnerability and makes the running process terminate abnormally.
After a program has crashed and terminated abnormally, it typically leaves behind a snapshot of its crashing state in the form of a core dump. While a core dump carries a large amount of information, which has long been used for software debugging, it barely serves as informative debugging aids in locating software faults, particularly memory corruption vulnerabilities. As such, previous research mainly seeks fully reproducible execution tracing to identify software vulnerabilities in crashes. However, such techniques are usually impractical for complex programs. Even for simple programs, the overhead of fully reproducible tracing may only be acceptable at the time of in-house testing.
In this talk, I will introduce a reverse execution technique, which takes as input a core dump, reversely executes the corresponding crashing program and automatically pinpoints the root cause of the vulnerable site hidden behind the crash. In the process of performing reverse execution, our technique typically encounters uncertainty (e.g., uncertain control or data flow) which significantly influence the capability of identifying vulnerabilities. Therefore, as part of the talk, I will also briefly discuss how we utilize deep recurrent neural network to tackle this technical challenge.
This lecture satisfies requirements for CSCI 591: Research Colloquium.
Biography: Dr. Xinyu Xing is an Assistant Professor at the Pennsylvania State University. His research interest includes exploring, designing and developing tools to automate vulnerability discovery, failure reproduction, vulnerability diagnosis (and triage), exploit and security patch generation. Recently, he is also interested in developing deep learning techniques to perform highly accurate binary and malware analysis. His past research has been featured by many mainstream medium, such as Technology Review, New Scientists and NYTimes etc. Going beyond academic research, he also actively participates and hosts many world-class cybersecurity competitions (such as HITB and XCTF). This year, his team was selected for DEFCON/GeekPwn CAAD challenge grand final at Las Vegas. He led Penn State to finish NSA code breaker competition 2017 and ranked at the top 3 nationwide. In the white-hat hacker community, his research team has contributed many CVEs for the open source community. The tools his team developed have been downloaded by thousands of developers and security researchers.
Host: Muhammad Naveed
Location: Henry Salvatori Computer Science Center (SAL) - 101
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Simone Silvetti and Laura Nenzi
Wed, Oct 10, 2018 @ 11:00 AM - 12:20 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Simone Silvetti (Numerical Methods Group) and Laura Nenzi (TU Wien),
Talk Title: Talk1: Combining Active learning optimization and Temporal logic for parameter synthesis and falsification of Complex Systems Talk 2: System Design of Stochastic Models using Robustness of Temporal Properties
Series: Computer Science Colloquium jointly with CCI-MHI Cyber-Physical Systems Seminar
Abstract: We are pleased to announce two talks during this colloquium.
Talk 1: Combining Active learning optimization and Temporal logic for parameter synthesis and falsification of Complex Systems
In this talk, we discuss the combination of Active Learning Optimization and temporal logic to the falsification and parameter synthesis of complex dynamical systems. First, we introduce Gaussian Processes and an active learning approach aimed to falsify a black box model with time-dependent functional inputs. Second, we introduce a technique also base on Gaussian Processes, named Smoothed Model Checking, which is able to estimate the probability that a stochastic system satisfies a temporal logic formula. We leverage this estimation ability and an active learning approach to find regions of the parameter space where the model satisfies a temporal logic formula with probability greater (or less) than a given threshold.
Talk 2: System Design of Stochastic Models using Robustness of Temporal Properties
In the last years, researchers from the verification community have proposed several notions of robustness for temporal logic providing suitable definitions of distance between a trajectory of a (deterministic) dynamical system and the boundaries of the set of trajectories satisfying the property of interest. In this talk, we present an extension of this notion of robustness to stochastic systems, showing that this naturally leads to a distribution of robustness degrees. Then, we show how to exploit this notion to address the system design problem, where the goal is to optimise some control parameters of a stochastic model in order to maximise robustness of the desired specifications. The key idea is to use a learning algorithm to estimate the dependence of the average robustness of a qualitative formula over the model parameters. A powerful and provably convergent machine learning method, namely the Gaussian Process - Upper Confidence Bound (GP-UCB) algorithm is use to improve the parameter optimisation. Finally, we demonstrate the applicability of our method on a number of case studies and ongoing works.
This lecture satisfies requirements for CSCI 591: Research Colloquium. Please note, due to limited capacity in OHE 100D, seats will be first come first serve.
Biography: Talk 1: Simone Silvetti is a researcher and developer at the Numerical Methods Group of Esteco SpA, Italy. He received a Ph.D. in Computer Science from the University of Udine in 2018 and a MSc in Mathematics from the University of Rome in 2012. His research focuses on the application of machine learning to quantitative formal methods and optimization.
Talk 2: Since 2017, Laura Nenzi is a research assistant at the TU Wien. She received a Ph.D in Computer Science from IMT Lucca, in 2016. In December 2018, she will join the University of Trieste as Assistant Professor. Her research interests include: spatio-temporal logics, statistical verification routines for uncertain models and combination of formal methods with machine learning techniques.
Host: Jyotirmoy Deshmukh and Paul Bogdan
Location: Olin Hall of Engineering (OHE) - 100D
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CAIS Seminar: Dr. Pinar Keskinocak (Georgia Tech) - Quantitative Models for Decision-Support in Healthcare Applications
Wed, Oct 10, 2018 @ 04:00 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Dr. Pinar Keskinocak, Georgia Institute of Technology
Talk Title: Quantitative Models for Decision-Support in Healthcare Applications
Series: USC Center for Artificial Intelligence in Society (CAIS) Seminar Series
Abstract: With the goal of improving patient outcomes, efficiency, and effectiveness, quantitative models are increasingly used for decision-support in healthcare. In this presentation Dr. Keskinocak will discuss a few applications from organ transplant, vaccination, screening, and workforce allocation decisions.
This lecture satisfies requirements for CSCI 591: Research Colloquium
Biography: Dr. Pinar Keskinocak is the William W. George Chair and Professor in the School of Industrial and Systems Engineering, and co-founder and Director of the Center for Health and Humanitarian Systems at Georgia Tech. She also serves as the College of Engineering ADVANCE Professor. Her research focuses on the applications of quantitative methods to have a positive impact in society, particularly in healthcare and humanitarian systems. She has worked on projects with a variety of governmental and non-governmental organizations, and healthcare providers, including American Red Cross, CDC, Children's Healthcare of Atlanta, Emory Healthcare, and Task Force for Global Health.
Host: Dr. Milind Tambe and Dr. Sze-chuan Suen
Location: Mark Taper Hall Of Humanities (THH) - 301
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Judea Pearl (UCLA) - The New Science of Cause and Effect
Mon, Oct 15, 2018 @ 12:00 PM - 01:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Professor Judea Pearl, UCLA
Talk Title: The New Science of Cause and Effect
Series: Computer Science Colloquium
Abstract: Professor Judea Pearl's talk will summarize a revolution that has changed the way scientists deal with cause-effect questions and that will have profound effects on our future. Pearl will first describe how hard causal questions that long were regarded as
either metaphysical or unmanageable can now be solved using elementary steps and what this tells us about how our mind achieves causal understanding. He will then outline how robots can be built that learn to communicate in our language of cause and effect and reason counterfactually about credit and blame, regret, intent and responsibility.
This lecture satisfies requirements for CSCI 591: Research Colloquium
This talk is co-sponsored by: USC Viterbi Department of Computer Science, USC Annenberg School for Communication and Journalism, USC Marshall Department of Data Sciences and Operations and USC Dornsife Department of Economics.
Biography: Judea Pearl is a professor of computer science and statistics at UCLA, where he directs the Cognitive Systems Laboratory and conducts research in artificial intelligence, human cognition, and philosophy of science. Pearl is the 2011 recipient of the ACM Alan Turing Award, and the 2002 London School of Economics Lakatos Award in the philosophy of science. He is the author of Heuristics (1983), Probabilistic Reasoning (1988), Causality (2000, 2009) and most recently co-authored The Book of Why (2018).
Host: Computer Science Department and School of Communication in Annenberg
Location: Wallis Annenberg Hall (ANN) 106 (Sheindlin Forum)
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Fred Morstatter (USC-ISI) - Discovering, Mitigating and Characterizing Social Data Bias
Wed, Oct 17, 2018 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Fred Morstatter, USC - ISI
Talk Title: Discovering, Mitigating and Characterizing Social Data Bias
Series: CS Colloquium
Abstract: Researchers and practitioners use social media to extract actionable patterns about human behavior. However, biases are inevitable and can either be a hindrance or an asset to such analysis. In this talk, I will discuss perturbations to the underlying data that can lead to flawed analysis. I will show how common assumptions in handling social media data can lead to flawed research results, and suggest approaches to combat these problems. However, if we understand the biases in our dataset it can lead to deeper understanding of the populations we wish to study. Once the biases underlying a social dataset are recognized, researchers are in a better position to study the unique characteristics underlying different cultural groups. This talk will conclude with a discussion of ways to identify cultural groups online and to characterize the biases between them.
This lecture satisfies requirements for CSCI 591: Research Colloquium. Please note, due to limited capacity, seats will be first come first serve.
Biography: Fred Morstatter is a Computer Scientist at the Information Science Institute. His research focuses on understanding biases that occur in online social data. Specifically, he is interested in biases that can skew research results from big social data. He is also interested in characterizing the biases of cultural groups based upon the trace data they create on social media. He has been a key contributor to the Synergistic Anticipation of Geopolitical Events (SAGE) project under IARPA's Hybrid Forecasting Competition. This project combines human judgement with machine forecasts of geopolitical events in the form of a web platform that serves as a vehicle for research in social media mining. He has published in JMLR, WWW, KDD and ICWSM, among others. He is Program Committee Chair for ICWSM 2019. A full list of publications can be found at www.fredmorstatter.com. Contact him at fredmors@isi.edu.
Host: Computer Science
Location: Olin Hall of Engineering (OHE) - 100D
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CAIS Seminar: Dr. John Prindle (USC) – Predicting Risk of Future Child Welfare Involvement
Wed, Oct 17, 2018 @ 04:00 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Dr. John Prindle, USC
Talk Title: Predicting Risk of Future Child Welfare Involvement
Series: USC Center for Artificial Intelligence in Society (CAIS) Seminar Series
Abstract: Child maltreatment impacts a significant number of children per year and is typically not limited to one encounter with the system. Past records provide a wealth of information which may be used to supplement current maltreatment allegations. Machine learning algorithms in the form of Random Forests were applied to these data to predict risk of future child welfare outcomes, past and present factors.
This lecture satisfies requirements for CSCI 591: Research Colloquium
Biography: Dr. John Prindle is a research assistant professor with the Children's Data Network at USC. His current work focuses on the impact of childhood maltreatment on downstream services such as education and medical services.
Host: Milind Tambe
Location: Mark Taper Hall Of Humanities (THH) - 301
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Steve Chien (JPL) - The Growing Role for Artificial Intelligence in Space Exploration and the Search for Life Beyond Earth
Wed, Oct 17, 2018 @ 05:00 PM - 06:30 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Steve Chien, Jet Propulsion Laboratory, California Institute of Technology
Talk Title: The Growing Role for Artificial Intelligence in Space Exploration and the Search for Life Beyond Earth
Series: Computer Science Colloquium
Abstract: Artificial Intelligence is playing an increasing role in our everyday lives and the business marketplace. This trend extends to the space sector, where AI has already shown considerable success and has the potential to revolutionize almost every aspect of space exploration. I first highlight a number of success stories of the tremendous impact of Artificial Intelligence in Space: over a dozen years of operations of the Autonomous Sciencecraft on EO-1, the Earth Observing Sensorweb tracking volcanoes, flooding and wildfires and automated targeting onboard the MSL Curiosity rover. Next I describe several search and optimization formulations of space scheduling problems: data management for spacecraft and observation scheduling. Finally I discuss why AI is critical to search for life beyond Earth, highlighting the role of AI in Europa Submersible and Interstellar mission concepts.
RSVP: https://goo.gl/forms/iLw0LrMKq6JvqxkD3
This lecture satisfies requirements for CSCI 591: Research Colloquium.
Biography: Dr. Steve Chien is a Senior Research Scientist at the Jet Propulsion Laboratory, California Institute of Technology where he leads efforts in autonomous systems for space exploration. Dr. Chien has received numerous awards for his research in space autonomous systems including: NASA Medals in 1997, 2000, 2007, and 2015; he is a four time honoree in the NASA Software of the Year competition (1999, 1999, 2005, 2011); and in 2011 he was awarded the inaugural AIAA Intelligent Systems Award. He has led the deployment of ground and flight autonomy software to numerous missions including the Autonomous Sciencecraft/Earth Observing One, WATCH/Mars Exploration Rovers, Earth Observing Sensorwebs, IPEX, and ESA's Rosetta.
Host: AAAI@USC
Location: Mark Taper Hall Of Humanities (THH) - 101
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Ioannis Mitliagkas (University of Montréal) - Negative Momentum for Improved Game Dynamics
Tue, Oct 23, 2018 @ 03:30 PM - 04:50 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Ioannis Mitliagkas, University of Montréal
Talk Title: Negative Momentum for Improved Game Dynamics
Series: Computer Science Colloquium
Abstract: Games generalize the single-objective optimization paradigm by introducing different objective functions for different players. Differentiable games often proceed by simultaneous or alternating gradient updates. In machine learning, games are gaining new importance through formulations like generative adversarial networks (GANs) and actor-critic systems. However, compared to single-objective optimization, game dynamics are more complex and less understood. In this talk, I will present recent research on the momentum dynamics of differentiable games. We will see an analysis of a simple differentiable game, which suggests that a negative momentum term can sometimes improve convergence. Then we will see empirical results that alternating gradient updates with a negative momentum term achieves convergence on the notoriously difficult to train saturating GANs.
This lecture satisfies requirements for CSCI 591: Research Colloquium.
Biography: Ioannis Mitliagkas is an assistant professor in the department of Computer Science and Operations Research (DIRO) at the University of Montréal, and member of MILA. Before that, he was a Postdoctoral Scholar with the departments of Statistics and Computer Science at Stanford University. He obtained his Ph.D. from the department of Electrical and Computer Engineering at The University of Texas at Austin. His research includes topics in statistical learning and inference, focusing on optimization, efficient large-scale and distributed algorithms, statistical learning theory and MCMC methods. His recent work includes methods for efficient and adaptive optimization, studying the interaction between optimization and the dynamics of large-scale learning systems as well as understanding and improving the performance of Gibbs samplers. In the past he has worked on high-dimensional streaming problems and fast algorithms and computation for large graph problems.
Host: Fei Sha
Location: Henry Salvatori Computer Science Center (SAL) - 101
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CAIS Seminar: Dr. Xiang Ren (USC) - Learning Text Structures with Weak Supervision
Wed, Oct 24, 2018 @ 04:00 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Dr. Xiang Ren, USC
Talk Title: Learning Text Structures with Weak Supervision
Series: USC Center for Artificial Intelligence in Society (CAIS) Seminar Series
Abstract: The real-world data, though massive, are hard for machines to resolve as they are largely unstructured and in the form of natural-language text. One of the grand challenges is to turn such massive corpora into machine-actionable structures. Yet, most existing systems have heavy reliance on human effort in the process of structuring various corpora, slowing down the development of downstream applications. In this talk, I will introduce an effort-light framework that extracts structured facts from massive corpora without task-specific human labeling effort. I will briefly introduce several interesting learning frameworks for structure extraction, and will share some directions towards mining corpus-specific structured networks for knowledge discovery.
This lecture satisfies requirements for CSCI 591: Research Colloquium
Biography: Xiang Ren is an Assistant Professor in the Department of Computer Science at USC affiliated with USC ISI. Xiang was a visiting researcher at Stanford University and received his PhD in CS at UIUC. He is interested in computational methods and systems that extract machine-actionable knowledge from massive unstructured text data, and is particularly excited about problems in the space of modeling sequence and graph data under weak supervision (learning with partial/noisy labels, and semi-supervised learning) and indirect supervision (multi-task learning, transfer learning, and reinforcement learning).
Location: Mark Taper Hall Of Humanities (THH) - 301
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Barna Saha (University of Massachusetts Amherst) - Efficient Fine-Grained Algorithms
Fri, Oct 26, 2018 @ 10:00 AM - 11:20 AM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Barna Saha, University of Massachusetts Amherst
Talk Title: Efficient Fine-Grained Algorithms
Series: Computer Science Colloquium
Abstract: One of the greatest successes of computational complexity theory is the classification of countless fundamental computational problems into polynomial-time and NP-hard ones, two classes that are often referred to as tractable and intractable, respectively. However, this crude distinction of algorithmic efficiency is clearly insufficient when handling today's large scale of data. We need a finer-grained design and analysis of algorithms that pinpoints the exact exponent of polynomial running time, and a better understanding of when a speed-up is not possible. Over the years, many polynomial-time approximation algorithms were devised as an approach to bypass the NP-hardness obstacle of many discrete optimization problems. This area has developed into a rich field producing many algorithmic ideas and has lead to major advances in computational complexity. So far, however, a similar theory for high polynomial time problems to understand the trade-off between quality and running time is vastly lacking.
In this presentation, I will give you an overview of the newly growing field of fine-grained algorithms and complexity, and my contributions therein. This will include fundamental problems such as edit distance computation, all-pairs shortest paths, parsing and matrix multiplication. They have applications ranging from genomics to statistical natural language processing, machine learning and optimization. I will show how as a natural byproduct of improved time complexity, one may design algorithms that are highly parallel as well as streaming algorithms with sublinear space complexity. Finally, motivated by core machine learning applications, I will discuss alternative measures of efficiency that may be equally relevant as time complexity.
This lecture satisfies requirements for CSCI 591: Research Colloquium.
Biography: Barna Saha is currently an Assistant Professor in the College of Information & Computer Science at the University of Massachusetts Amherst. She is also a Permanent Member of Center for Discrete Mathematics and Theoretical Computer Science (DIMACS) at Rutgers. Before joining UMass in 2014, she was a Research Scientist at AT&T Shannon Laboratories, New Jersey. She spent four wonderful years (2007-2011) at the University of Maryland College Park from where she received her Ph.D. in Computer Science. In Fall 2015, she was at the University of California Berkeley as a Visiting Scholar and as a fellow of the Simons Institute. Her research interests include Theoretical Computer Science, Probabilistic Method & Randomized Algorithms and Large Scale Data Analytics. She is the recipient of NSF CAREER award (2017), Google Faculty Award (2016), Yahoo Academic Career Enhancement Award (2015), Simons-Berkeley Research Fellowship (2015), NSF CRII Award (2015) and Dean's Dissertation Fellowship (2011). She received the best paper award at the Very Large Data Bases Conference (VLDB) 2009 for her work on Probabilistic Databases and was chosen as finalists for best papers at the IEEE Conference on Data Engineering (ICDE) 2012 for developing new techniques to handle low quality data.
Host: David Kempe
Location: Seeley G. Mudd Building (SGM) - 101
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Distinguished Lecture: Nina Balcan (CMU) - Data Driven Algorithm Design
Tue, Oct 30, 2018 @ 03:30 PM - 04:50 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Nina Balcan, Carnegie Mellon University
Talk Title: Data Driven Algorithm Design
Series: Computer Science Distinguished Lecture Series
Abstract: Data driven algorithm design for combinatorial problems is an important aspect of modern data science and algorithm design. Rather than using off the shelf algorithms that only have worst case performance guarantees, practitioners typically optimize over large families of parametrized algorithms and tune the parameters of these algorithms using a training set of problem instances from their domain to determine a configuration with high expected performance over future instances. However, most of this work comes with no performance guarantees. The challenge is that for many combinatorial problems, including partitioning and subset selection problems, a small tweak to the parameters can cause a cascade of changes in the algorithm's behavior, so the algorithm's performance is a discontinuous function of its parameters.
In this talk, I will present new work that helps put data driven combinatorial algorithm selection on firm foundations. We provide strong computational and statistical performance guarantees for several subset selection and combinatorial partitioning problems (including various forms of clustering), both for the batch and online scenarios where a collection of typical problem instances from the given application are presented either all at once or in an online fashion, respectively.
This lecture satisfies requirements for CSCI 591: Research Colloquium.
Biography: Maria Florina Balcan is an Associate Professor in the School of Computer Science at Carnegie Mellon University. Her main research interests are machine learning, computational aspects in economics and game theory, and algorithms. Her honors include the CMU SCS Distinguished Dissertation Award, an NSF CAREER Award, a Microsoft Faculty Research Fellowship, a Sloan Research Fellowship, and several paper awards. She was a program committee co-chair for the Conference on Learning Theory in 2014 and for the International Conference on Machine Learning in 2016. She is currently board member of the International Machine Learning Society (since 2011), a Tutorial Chair for ICML 2019, and a Workshop Chair for FOCS 2019.
Host: Computer Science Department
Location: Henry Salvatori Computer Science Center (SAL) - 101
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Tech Talk: AI for Content Creation and Interaction
Wed, Oct 31, 2018 @ 04:00 PM - 06:20 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Dr. Lei Li & Dr. Jianchao Yang, ByteDance AI Lab
Talk Title: AI for Content Creation and Interaction
Abstract: In the mobile era, we are being presented an exciting opportunity to shape the way people acquire and consume information. In this talk, we will reveal the roles of AI technologies in the information consumption platforms. We will share several recent work at ByteDance AI Lab towards more efficient creation of and interaction with content. We will introduce a robot writer, Xiaomingbot, which has produced more than 60k articles since August 2016, some of them in multiple languages including English, Chinese and Portuguese. It relies on state-of-the-art representation learning for sentences and generative models from data, text, and images. We will also introduce our latest research in visual understanding of objects and scene in short videos, and how these technologies assist authors to create better content. The talk will be accompanied with interactive demos of these technologies in Tiktok(Douyin), Vigo(Huoshan), and Toutiao apps.
Biography: Dr. Lei Li is Director of ByteDance AI Lab. Lei received his B.S. in Computer Science and Engineering from Shanghai Jiao Tong University (ACM class) and Ph.D. in Computer Science from Carnegie Mellon University, respectively. His dissertation work on fast algorithms for mining co-evolving time series was awarded ACM KDD best dissertation (runner up). His recent work on AI writing received 2nd-class award of WU Wenjun AI prize of China. Before Toutiao, he worked at Baidu's Institute of Deep Learning in Silicon Valley as a Principal Research Scientist. Before that, he was working in EECS department of UC Berkeley as a Post-Doctoral Researcher. He has served in the Program Committee for ICML 2014, ECML/PKDD 2014/2015, SDM 2013/2014, IJCAI 2011/2013/2016, KDD 2015/2016, 2017 KDD Cup co-Chair, KDD 2018 hands-on tutorial co-chair, and as a lecturer in 2014 summer school on Probabilistic Programming for Advancing Machine Learning. He has published over 40 technical papers and holds 3 US patents.
Dr. Jianchao Yang is Director of ByteDance AI Lab US. Before joining ByteDance, Jianchao was a manager and principal research scientist at Snap, where he led the computer vision area. He obtained his Ph.D. degree under supervision of Prof. Thomas Huang from University of Illinois at Urbana-Champaign. He has published over 80 technical papers on top conferences and journals, which have attracted over 15k citations from the research community. He is the receipt of Best Student Paper Award in ICCV 2011. He and his collaborators are multiple winners of international competitions and challenges, including PASCAL VOC 2009, ImageNet 2014, WebVision 2017, and NTIRE Super-resolution Challenge 2018.
Host: Xiang Ren
Location: Grace Ford Salvatori Hall Of Letters, Arts & Sciences (GFS) - 116
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
