Select a calendar:
Filter March Events by Event Type:
SUNMONTUEWEDTHUFRISAT
Conferences, Lectures, & Seminars
Events for March
-
Taylor Berg-Kirkpatrick (Carnegie Mellon University) – Balancing Constraint and Flexibility in Unsupervised Models for Language Analysis
Thu, Mar 01, 2018 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Taylor Berg-Kirkpatrick, Carnegie Mellon University
Talk Title: Balancing Constraint and Flexibility in Unsupervised Models for Language Analysis
Series: Computer Science Colloquium
Abstract: Without careful consideration of the relationship between input and output, unsupervised learning problems can be under-constrained. This talk will discuss approaches for making unsupervised problems feasible by incorporating different types of inductive bias. First, we focus on a set of raw data analysis tasks related to the digital humanities, including historical document recognition, music transcription, and compositor attribution. For each of these tasks, strong prior knowledge about the causal process behind the data can be encoded into the model. We show how to leverage this casual knowledge as a helpful source of constraint, yielding systems that in some cases outperform their supervised counterparts. Next, we investigate several linguistic analysis tasks where causal structure is more difficult to encode. Here, we develop a new unsupervised model class that combines structured and continuous representations by leveraging the flexibility of neural networks. We show that incorporating a volume-preserving constraint on the neural component of our model makes learning well-behaved. Using this approach, we demonstrate start-of-the-art results on two standard unsupervised NLP tasks: part-of-speech induction and unsupervised dependency parsing.
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: Taylor Berg-Kirkpatrick joined the Language Technologies Institute at Carnegie Mellon University as an Assistant Professor in Fall 2016. Previously, he was a Research Scientist at Semantic Machines Inc. and, before that, completed his Ph.D. in computer science at the University of California, Berkeley. Taylor's research focuses on using machine learning to understand structured human data, including language but also sources like music, document images, and other complex artifacts.
Host: Computer Science Department
Location: Olin Hall of Engineering (OHE) - 100D
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: TBA
Tue, Mar 06, 2018 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: TBA, TBA
Talk Title: TBA
Series: CS Colloquium
Abstract: TBA
This lecture satisfies requirements for CSCI 591: Research Colloquium. Please note, due to limited capacity, seats will be first come first serve.
Biography: TBA
Host: Muhammad Naveed / David Kempe
Location: Olin Hall of Engineering (OHE) - 100D
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CAIS Seminar: Andrew Perrault (University of Toronto) – Developing and Coordinating Autonomous Agents for Efficient Electricity Markets
Wed, Mar 07, 2018 @ 04:00 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Andrew Perrault, University of Toronto
Talk Title: Developing and Coordinating Autonomous Agents for Efficient Electricity Markets
Series: USC Center for Artificial Intelligence in Society (CAIS) Seminar Series
Abstract: Aggressive greenhouse gas reduction targets will necessitate a transformation of energy use systems, with increasing emphasis on electricity, which can be decarbonized more efficiently than other energy sources. Mr. Perrault argues that deploying consumer-representing autonomous agents can make this transformation less expensive by allowing attention-limited consumers to respond to changes in market conditions. The talk has two parts: in Part I, he develops a cooperative game theoretic model that illustrates the value of such agents in electricity markets. In Part II, he focuses on the problem of training such an agent using a new variant of preference elicitation called experiential elicitation.
This lecture satisfies requirements for CSCI 591: Research Colloquium
Biography: Andrew Perrault is a PhD student at University of Toronto, supervised by Craig Boutilier. His research focuses on the application of AI to electricity markets and electricity use. He is the co-founder and co-lead developer at theschoolfund.org, a non-profit that crowdfunds scholarships for secondary school students in developing countries.
Host: Milind Tambe
Location: Seeley Wintersmith Mudd Memorial Hall (of Philosophy) (MHP) - 101
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Jonas Mueller (MIT) – Learning Optimal Interventions under Uncertainty
Thu, Mar 08, 2018 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Jonas Mueller, MIT
Talk Title: Learning Optimal Interventions under Uncertainty
Series: Computer Science Colloquium
Abstract: A basic goal of data analysis is learning which actions (ie. interventions) are best for producing desired outcomes. While advances in reinforcement learning and bandit/Bayesian optimization have shown great promise, these sequential methods are primarily limited to digital environments where iterating between modeling and experimentation is easy. Although more widely applicable, learning from a fixed (observational) dataset will inherently involve substantial uncertainty due to limited samples, and it is undesirable to prescribe actions whose outcomes are unclear.
In this talk, I will consider such settings from a Bayesian perspective and formalize the of role of uncertainty in data-driven actions. Adopting a Gaussian process framework, I will introduce a conservative definition of the optimal intervention which can be either tailored on an individual basis or globally enacted over a population. Subsequently, these ideas are extended to structured sequence data via a recurrent variational autoencoder model. In both cases, gradient methods are employed to identify the best intervention and a key theme of the approach is carefully constraining this optimization to avoid regions of high uncertainty. Various applications of this methodology will presented including gene expression manipulation, therapeutic antibody design, and revision of natural language.
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: Jonas Mueller is a Computer Science Ph.D. student at MIT working with Tommi Jaakkola and David Gifford. His research interests lie in developing machine learning methods to advance both statistical science and artificial intelligence applications. Integrating ideas from optimal transport, deep learning, Bayesian/bandit optimization, and interpretable modeling, much of his work has been motivated by applications in bioinformatics and natural language processing. Previously, Jonas studied Math and Statistics at UC Berkeley, where he was awarded the Departmental Citation, and he recently also spent some time at Microsoft Research.
Host: Computer Science Department
Location: Olin Hall of Engineering (OHE) - 100D
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
TBA
Tue, Mar 20, 2018 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: TBA,
Talk Title: TBA
Series: CS Colloquium
Abstract: TBA
This lecture satisfies requirements for CSCI 591: Research Colloquium. Please note, due to limited capacity, seats will be first come first serve.
Biography: TBA
Host: Muhammad Naveed / David Kempe
Location: Olin Hall of Engineering (OHE) - 100 D
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CS Distinguished Lecture: Sham Kakade (University of Washington) – Sub-Linear Reinforcement Learning
Tue, Mar 20, 2018 @ 04:00 PM - 05:20 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Sham Kakade, University of Washington
Talk Title: Sub-Linear Reinforcement Learning
Series: Computer Science Distinguished Lecture Series
Abstract: Suppose an agent is an unknown environment and seeks to maximize his/her long term future reward. We consider the basic question: does the agent need to learn an accurate model of the environment before he/she can start executing a near-optimal long term course of actions?
Specifically, this talk will consider the problem of provably optimal reinforcement learning for (episodic) finite horizon MDPs, i.e., how an agent learns to maximize his/her (long term) reward in an uncertain environment. The talk will present a novel algorithm, the Variance-reduced Upper Confidence Q-learning (vUCQ), which is the first algorithm which enjoys a regret bound that is both sub-linear in the model size and that achieves optimal minimax regret. The algorithm is sub-linear in that the time to achieve epsilon average regret is a number of samples that is far less than that required to learn any (non-trivial) estimate of the underlying model of the environment. The importance of sub-linear algorithms is largely the motivation for algorithms such as "Q-learning" and other "model-free" approaches.
vUCQ is a successive refinement method in which the algorithm reduces the variance in the "Q-value" estimates and couples this estimation scheme with an upper confidence based algorithm. Technically, this coupling of these techniques is what leads to the algorithm's strong guarantees, showing that "model-free" approaches can be optimal.
This lecture satisfies requirements for CSCI 591: Research Colloquium.
Biography: Sham Kakade is a Washington Research Foundation Data Science Chair, with a joint appointment in the Department of Statistics and the Department of Computer Science at the University of Washington.
From 2011-2015, I was a principal research scientist at Microsoft Research, New England. From 2010-2012, I was an associate professor at the Department of Statistics, Wharton, University of Pennsylvania. From 2005-2009, I was an assistant professor at the Toyota Technological Institute at Chicago.
I completed my PhD at the Gatsby Computational Neuroscience Unit under the supervision of Peter Dayan, and I was an undergraduate at Caltech where I obtained my BS in physics. I was a postdoc in the Computer and Information Science department at the University of Pennsylvania under the supervision of Michael Kearns.
Host: Computer Science Department
Location: Henry Salvatori Computer Science Center (SAL) - 101
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Nithya Sambasivan (Google) - Design for Autonomy and Fairness of New Technology Users in the Global South
Wed, Mar 21, 2018 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Nithya Sambasivan, Google
Talk Title: Design for Autonomy and Fairness of New Technology Users in the Global South
Series: CS Colloquium
Abstract: 2017 saw half the world online. As technology penetration and ecosystem maturity increase, there is a growing intent to use technology for socio-economic development for new technology users. However, complex long-standing challenges like affordability, safety, and socio-religious diktats affect people at the cusp of the internet. My work aims to empower new technology users with increased autonomy and fairness through technology. I present my prior work on design and evaluation of a cost transparency tool intended to help new mobile Internet users; design to tackle abuse and safety vectors for women in Internet technologies; and design and deployment of an information broadcasting system for urban sex workers in India. I show how prevailing HCI assumptions of privacy, trust, and user identities need to be challenged as Internet advances to reach all edges of human society. Through these projects, I show how large problems can be practically addressed through a combination of design, policy, and algorithms.
This lecture satisfies requirements for CSCI 591: Research Colloquium. Please note, due to limited capacity, seats will be first come first serve.
Biography: Nithya Sambasivan is a researcher focused on technology design for social, economic and political benefits in the Global South. Her research spans the areas of HCI and ICTD, and has won several recognitions at top conferences. She has been a researcher at Google since 2012, where she has co-founded a group to conduct future-facing research on under-represented topics, such as gender equity and new technology users. Her research has influenced several large-scale real-world projects for the next billion users, and has been directly translated to core libraries, metrics, and guidance for Android and Web developers at Build for Billions, design.google/nbu, and Google I/O talks. Nithya has a Ph.D. and MS in information and computer sciences for University of California, Irvine and and MS in Human Computer Interaction (HCI) from Georgia Tech. Her dissertation focused on technology design for the low-income communities of slums, urban sex workers and microentreprises in India. She is a recipient of Google's Anita Borg and UC Irvine Dean's fellowships. She has interned at Microsoft Research India, Nokia Research Center and IBM TJ Watson.
Host: Milind Tambe
Location: Ronald Tutor Hall of Engineering (RTH) - 115
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CAIS Seminar: Dr. Ian Holloway (UCLA) - Social Networking Site Data Mining to Understand Substance Use and HIV Risk Among Gay, Bisexual and Other Men Who Have Sex With Men
Wed, Mar 21, 2018 @ 04:00 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Dr. Ian Holloway, UCLA
Talk Title: Social Networking Site Data Mining to Understand Substance Use and HIV Risk Among Gay, Bisexual and Other Men Who Have Sex With Men
Series: USC Center for Artificial Intelligence in Society (CAIS) Seminar Series
Abstract: Dr. Holloway's presentation will outline the development of a culturally congruent data collection and mining module (DCMM) to study the social networking site (SNS) use patterns, substance use and HIV risk and protective behaviors of gay, bisexual and other men who have sex with men (MSM). Data gathered through the DCMM will be used to inform just-in-time adaptive interventions to prevent incidence of new HIV cases among this population disproportionately impacted by HIV/AIDS.
This lecture satisfies requirements for CSCI 591: Research Colloquium
Biography: Dr. Holloway is an Assistant Professor in the Department of Welfare at the UCLA Luskin School of Public Affairs and the Director of the Southern California HIV/AIDS Policy Research Center. His applied behavioral health research examines the contextual factors that contribute to heath disparities among sexual and gender minority populations. Dr. Holloway is particularly interested in how social media and new technologies can be harnessed for health promotion and disease prevention. He holds dual master's degrees in social work and public health from Columbia University and a doctorate in social work from the University of Southern California.
Host: Milind Tambe
Location: Seeley Wintersmith Mudd Memorial Hall (of Philosophy) (MHP) - 101
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Mark Bun (Princeton University) - Finding Structure in the Landscape of Differential Privacy
Wed, Mar 21, 2018 @ 04:00 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Mark Bun, Princeton University
Talk Title: Finding Structure in the Landscape of Differential Privacy
Series: CS Colloquium
Abstract: Differential privacy offers a mathematical framework for balancing two goals: obtaining useful information about sensitive data, and protecting individual-level privacy. Discovering the limitations of differential privacy yields insights as to what analyses are incompatible with privacy and why. These insights further aid the quest to discover optimal privacy-preserving algorithms. In this talk, I will give examples of how both follow from new understandings of the structure of differential privacy.
I will first describe negative results for private data analysis via a connection to cryptographic objects called fingerprinting codes. These results show that an (asymptotically) optimal way to solve natural high-dimensional tasks is to decompose them into many simpler tasks. In the second part of the talk, I will discuss concentrated differential privacy, a framework which enables more accurate analyses by precisely capturing how simpler tasks compose
This lecture satisfies requirements for CSCI 591: Research Colloquium. Please note, due to limited capacity, seats will be first come first serve.
Biography: Mark Bun is a postdoctoral researcher in the Computer Science Department at Princeton University. He is broadly interested in theoretical computer science, and his research focuses on understanding foundational problems in data privacy through the lens of computational complexity theory. He completed his Ph.D. at Harvard in 2016, where he was advised by Salil Vadhan and supported by an NDSEG Research Fellowship.
Host: Aleksandra Korolova
Location: Ronald Tutor Hall of Engineering (RTH) - 217
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CS Colloquium: Holly Yanco (University of Massachusetts Lowell) - Designing for Human-Robot Interaction
Thu, Mar 22, 2018 @ 04:00 PM - 05:20 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Holly Yanco, University of Massachusetts Lowell
Talk Title: Designing for Human-Robot Interaction
Series: Computer Science Colloquium
Abstract: Robots navigating in difficult and dynamic environments often need assistance from human operators or supervisors, either in the form of teleoperation or interventions when the robot's autonomy is not able to handle the current situation. Even in more controlled environments, such as office buildings and manufacturing floors, robots may need help from people. This talk will discuss methods for controlling both individual robots and groups of robots, in applications ranging from assistive technology to telepresence to search and rescue. A variety of modalities for human-robot interaction with robot systems, including multi-touch devices, software-based operator control units (softOCUs), game controllers, virtual reality headsets, and Google Glass, will be presented.
This lecture satisfies requirements for CSCI 591: Research Colloquium.
Biography: Dr. Holly Yanco is a Distinguished University Professor, Professor of Computer Science, and Director of the New England Robotics Validation and Experimentation (NERVE) Center at the University of Massachusetts Lowell. Her research interests include human-robot interaction, multi-touch computing, robot autonomy, fostering trust of autonomous systems, evaluation methods for robot systems, and the use of robots in K-12 education to broaden participation in computer science. Yanco's research has been funded by NSF, including a CAREER Award, ARO, DARPA, DOE-EM, NASA, NIST, Microsoft, and Google. Yanco is Co-Chair of the Massachusetts Technology Leadership Council's Robotics Cluster,served as Co-Chair of the Steering Committee for the ACM/IEEE Conference on Human-Robot Interaction and Journal of Human-Robot Interaction from 2013-2016, and was a member of the Executive Council of the Association for the Advancement of Artificial Intelligence (AAAI) from 2006-2009. Yanco has a PhD in Computer Science from the Massachusetts Institute of Technology.
Host: Maja Mataric
Location: Henry Salvatori Computer Science Center (SAL) - 101
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Himabindu Lakkaraju (Stanford University) Human-Centric Machine Learning: Enabling Machine Learning for High-Stakes Decision-Making
Mon, Mar 26, 2018 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Himabindu Lakkaraju, Stanford University
Talk Title: Human-Centric Machine Learning: Enabling Machine Learning for High-Stakes Decision-Making
Series: CS Colloquium
Abstract: Domains such as law, healthcare, and public policy often involve highly consequential decisions which are predominantly made by human decision-makers. The growing availability of data pertaining to such decisions offers an unprecedented opportunity to develop machine learning models which can aid human decision-makers in making better decisions. However, the applicability of machine learning to the aforementioned domains is limited by certain fundamental challenges:
1) The data is selectively labeled i.e., we only observe the outcomes of the decisions made by human decision-makers and not the counterfactuals.
2) The data is prone to a variety of selection biases and confounding effects.
3) The successful adoption of the models that we develop depends on how well decision-makers can understand and trust their functionality, however, most of the existing machine learning models are primarily optimized for predictive accuracy and are not very interpretable.
In this talk, I will describe novel computational frameworks which address the aforementioned challenges, thus, paving the way for large-scale deployment of machine learning models to address problems of significant societal impact. First, I will discuss how to build interpretable predictive models and explanations of complex black box models which can be readily understood and consequently trusted by human decision-makers. I will then outline efficient and provably near-optimal approximation algorithms to solve these problems. Next, I will present a novel evaluation framework which allows us to reliably compare the quality of decisions made by human decision-makers and machine learning models amidst challenges such as missing counterfactuals and presence of unmeasured confounders (unobservables). Lastly, I will provide a brief overview of my research on diagnosing and characterizing biases (systematic errors) in human decisions and predictions of machine learning models.
I will conclude the talk by sketching future directions which enable effective and efficient collaboration between humans and machine learning models to address problems of societal impact.
This lecture satisfies requirements for CSCI 591: Research Colloquium. Please note, due to limited capacity, seats will be first come first serve.
Biography: Hima Lakkaraju is a Ph.D. candidate in Computer Science at Stanford University. Her research focuses on enabling machine learning models to complement human decision making in high-stakes settings such as law, healthcare, public policy, and education. At the core of her research lie rigorous computational techniques leading to algorithmic contributions in machine learning, data mining, and econometrics. Hima has received several fellowships and awards including the Robert Bosch Stanford graduate fellowship, Microsoft research dissertation grant, Google Anita Borg scholarship, IBM eminence and excellence award, and best paper awards at SIAM International Conference on Data Mining (SDM) and INFORMS. Her research has been covered by various media outlets such as the New York Times, MIT Tech Review, Harvard Business Review, TIME, Forbes, Business Insider, and Bloomberg.
Host: Aleksandra Korolova
Location: Ronald Tutor Hall of Engineering (RTH) - 115
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CS Colloquium: Stefanos Nikolaidis (Carnegie Mellon University) - Mathematical Models of Adaptation in Human-Robot Collaboration
Tue, Mar 27, 2018 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Stefanos Nikolaidis, Carnegie Mellon University
Talk Title: Mathematical Models of Adaptation in Human-Robot Collaboration
Series: CS Colloquium
Abstract: The goal of my research is to improve human-robot collaboration by integrating mathematical models of human behavior into robot decision making. I develop game-theoretic algorithms and probabilistic planning techniques that reason over the uncertainty in the human internal state and its dynamics, enabling autonomous systems to act optimally in a variety of real-world collaborative settings.
While much work in human-robot interaction has focused on leader-assistant teamwork models, the recent advancement of robotic systems that have access to vast amounts of information suggests the need for robots that take into account the quality of the human decision making and actively guide people towards better ways of doing their task. In this talk, I propose an equal partners model, where human and robot engage in a dance of inference and action, and I focus on one particular instance of this dance: the robot adapts its own actions via estimating the probability of the human adapting to the robot. I start with a bounded memory model of human adaptation parameterized by the human adaptability - the probability of the human switching towards a strategy newly demonstrated by the robot. I then propose data-driven models that capture subtler forms of adaptation, where the human teammate updates their expectations of the robot's capabilities through interaction. Integrating these models into robot decision making allows for human-robot mutual adaptation, where coordination strategies, informative actions and trustworthy behavior are not explicitly modeled, but naturally emerge out of optimization processes. Human subjects experiments in a variety of collaboration and shared autonomy settings show that mutual adaptation significantly improves human-robot team performance, compared to one-way robot adaptation to the human.
This lecture satisfies requirements for CSCI 591: Research Colloquium. Please note, due to limited capacity, seats will be first come first serve.
Biography: Stefanos Nikolaidis completed his PhD at Carnegie Mellon's Robotics Institute in December 2017 and he is currently a research associate at the University of Washington, Computer Science & Engineering. His research lies at the intersection of human-robot interaction, algorithmic game-theory and planning under uncertainty. Stefanos develops decision making algorithms that leverage mathematical models of human behavior to support deployed robotic systems in real-world collaborative settings. He has a MS from MIT, a MEng from the University of Tokyo and a BS from the National Technical University of Athens. He has additionally worked as a research specialist at MIT and as a researcher at Square Enix in Tokyo. He has received a Best Enabling Technologies Paper Award from the IEEE/ACM International Conference on Human-Robot Interaction, has a best paper nomination from the same conference this year and was a best paper award finalist in the International Symposium on Robotics.
Host: Joseph Lim
Location: Olin Hall of Engineering (OHE) - 100D
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CAIS Seminar: Dr. Mayank Kejriwal (USC Information Sciences Institute) - Building Knowledge Graphs for Social Good
Wed, Mar 28, 2018 @ 04:00 PM - 05:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Dr. Mayank Kejriwal, USC Information Sciences Institute
Talk Title: Building Knowledge Graphs for Social Good
Series: USC Center for Artificial Intelligence in Society (CAIS) Seminar Series
Abstract: Illicit activities like human trafficking and narcotics have a significant Web footprint. In this talk, I will introduce and talk about building knowledge graphs (KG), a powerful means of representing and reasoning over knowledge using intelligent algorithms, to combat such problems for social good. I will also introduce a KG-centric system called DIG, developed in our group, that is currently being used by more than 100 US law enforcement agencies to combat human trafficking.
This lecture satisfies requirements for CSCI 591: Research Colloquium
Biography: Dr. Mayank Kejriwal is a researcher at the USC Information Sciences Institute. His research on knowledge graphs, currently funded under both DARPA and IARPA, has been published in multiple interdisciplinary ACM, IEEE, Springer and Elsevier venues. He is authoring a textbook on knowledge graphs (MIT Press) with Pedro Szekely and Craig Knoblock.
Host: Milind Tambe
Location: Mark Taper Hall Of Humanities (THH) - 102
Audiences: Everyone Is Invited
Contact: Computer Science Department
-
CS Colloquium: Junier Oliva (Carnegie Mellon University) Scalable Learning Over Distributions
Thu, Mar 29, 2018 @ 11:00 AM - 12:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Junier Oliva, Carnegie Mellon University
Talk Title: Scalable Learning Over Distributions
Series: CS Colloquium
Abstract: A great deal of attention has been applied to studying new and better ways to perform learning tasks involving static finite vectors. Indeed, over the past century the fields of statistics and machine learning have amassed a vast understanding of various learning tasks like clustering, classification, and regression using simple real valued vectors. However, we do not live in a world of simple objects. From the contact lists we keep, the sound waves we hear, and the distribution of cells we have, complex objects such as sets, distributions, sequences, and functions are all around us. Furthermore, with ever-increasing data collection capacities at our disposal, not only are we collecting more data, but richer and more bountiful complex data are becoming the norm.
In this presentation we analyze regression problems where input covariates, and possibly output responses, are probability distribution functions from a nonparametric function class. Such problems cover a large range of interesting applications including learning the dynamics of cosmological particles and general tasks like parameter estimation.
However, previous nonparametric estimators for functional regression problems scale badly computationally with the number of input/output pairs in a data-set. Yet, given the complexity of distributional data it may be necessary to consider large data-sets in order to achieve a low estimation risk.
To address this issue, we present two novel scalable nonparametric estimators: the Double-Basis Estimator (2BE) for distribution-to-real regression problems; and the Triple-Basis Estimator (3BE) for distribution-to-distribution regression problems. Both the 2BE and 3BE can scale to massive data-sets. We show an improvement of several orders of magnitude in terms of prediction speed and a reduction in error over previous estimators in various synthetic and real-world data-sets.
This lecture satisfies requirements for CSCI 591: Research Colloquium. Please note, due to limited capacity, seats will be first come first serve.
Biography: Junier Oliva is a Ph.D. candidate in the Machine Learning Department at the School of Computer Science, Carnegie Mellon University. His main research interest is to build algorithms that understand data at an aggregate, holistic level. Currently, he is working to push machine learning past the realm of operating over static finite vectors, and start reasoning ubiquitously with complex, dynamic collections like sets and sequences. Moreover, he is interested in exporting concepts from learning on distributional and functional inputs to modern techniques in deep learning, and vice-versa. He is also developing methods for analyzing massive datasets, both in terms of instances and covariates. Prior to beginning his Ph.D. program, he received his B.S. and M.S. in Computer Science from Carnegie Mellon University. He also spent a year as a software engineer for Yahoo!, and a summer as a machine learning intern at Uber ATG.
Host: Fei Sha
Location: Olin Hall of Engineering (OHE) - 100D
Audiences: Everyone Is Invited
Contact: Assistant to CS chair
-
CS Distinguished Lecture: Satinder Singh (University of Michigan) – Reinforcement Learning: From Vision to Action and Back
Thu, Mar 29, 2018 @ 04:00 PM - 05:20 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Satinder Singh, University of Michigan
Talk Title: Reinforcement Learning: From Vision to Action and Back
Series: Computer Science Distinguished Lecture Series
Abstract: Stemming in part from the great successes of other areas of Machine Learning, in particular the recent success of Deep Learning, there is renewed hope and interest in Reinforcement Learning (RL) from the wider applications communities. Indeed, there is a recent burst of new and exciting progress in both theory and practice of RL. I will describe some theoretical results from my own group on a simple new connection between planning horizon and overfitting in RL, as well as some results on combining RL with Deep Learning in Minecraft, and Zero-Shot Generalization across compositional tasks. I will conclude with some lookahead at what we can do, both as theoreticians and those that collect data, to accelerate the impact of RL.
This lecture satisfies requirements for CSCI 591: Research Colloquium.
Biography: Satinder Singh is a Professor of Computer Science and Engineering at the University of Michigan, Ann Arbor. He has been the Chief Scientist at Syntek Capital, a venture capital company, a Principal Research Scientist at AT&T Labs, an Assistant Professor of Computer Science at the University of Colorado, Boulder, and a Postdoctoral Fellow at MIT's Brain and Cognitive Science department. His research focus is on developing the theory, algorithms and practice of building artificial agents that can learn from interaction in complex, dynamic, and uncertain environments, including environments with other agents in them. His main contributions have been to the areas of reinforcement learning, multi-agent learning, and more recently to applications in cognitive science and healthcare. He is a Fellow of the AAAI (Association for the Advancement of Artificial Intelligence) and has coauthored more than 150 refereed papers in journals and conferences and has served on many program committee's. He was Program-CoChair of AAAI 2017, and in 2013 helped cofound RLDM (Reinforcement Learning and Decision Making), a biennial multidisciplinary meeting that brings together computer scientists, psychologists, neuroscientists, roboticists, control theorists, and others interested in animal and artificial decision making.
Host: Haipeng Luo
Location: Henry Salvatori Computer Science Center (SAL) - 101
Audiences: Everyone Is Invited
Contact: Computer Science Department
