Event Archive

Policy gradient methods for reinforcement learning and continuous control are popular in practice and have helped recent advances in robotic navigation and in game playing. However, they lack theoretical guarantees even for the simplest case of linear dynamics and a quadratic cost, the Linear Quadratic Regulator (LQR) problem. A difficulty is that unlike the classical approaches to LQR, these methods must solve a nonconvex optimization problem to find the optimal control policy.

We study the problem of fairly allocating a set of indivisible items among $n$ agents. Typically, the literature has focused on one-shot algorithms. In this talk we depart from this paradigm and allow items to arrive online. When an item arrives we must immediately and irrevocably allocate it to an agent. A paradigmatic example is that of food banks: food donations arrive, and must be delivered to nonprofit organizations such as food pantries and soup kitchens.

When you think of McDonald’s, the Golden Arches, a Big Mac, or a favorite Happy Meal memory likely spring to mind; but you might not immediately think of the McDonald’s App.  The US Digital team is working to change that.  McDonald’s is transforming their business, driven by digital.  From revolutionizing ordering to personalizing offers, the US Digital team is using data and technology to improve the customer experience and drive sales growth.  Come listen to Hashim Amin, Head of Digital – US Market, speak about McDonald’s digital journey.

The Graph Matching problem is a robust version of the Graph Isomorphism problem: given two not-necessarily-isomorphic graphs, the goal is to find a permutation of the vertices which maximizes the number of common edges. We study a popular average-case variant; we deviate from the common heuristic strategy and give the first quasi-polynomial time algorithm, where previously only sub-exponential time algorithms were known. 

Today, we are collecting an immense amount of health data both inside and outside of the hospital. While clinicians are studying ever more data about their patients, they are still ignoring the vast majority of it. Transforming these observational data into actionable knowledge is challenging due to a number of reasons including the presence of confounders, missing context, and complex longitudinal relationships. At the same time, due to the high-stakes nature of healthcare, the field requires tools that are not only accurate, but also interpretable and robust.

The most prevalent notions of fairness in machine learning are statistical definitions: they fix a small collection of pre-defined attributes (such as race, gender, age or disability), and then ask for parity of some statistic of the classifier across these attributes. Constraints of this form are susceptible to intentional or inadvertent "fairness gerrymandering", in which a classifier appears to be fair on each attribute marginally, but badly violates the fairness constraint on one or more structured subgroups (such as disabled Hispanic women over age 55).

Advances in computational optimization allow for the organization of large combinatorial markets. We aim for allocations and competitive equilibrium prices, i.e. outcomes that are in the core. The research is motivated by the design of environmental markets, but similar problems appear in energy and logistics markets or in the allocation of airport time slots. Budget constraints are an important concern in many of these markets.

Dr. Sandra K. Johnson is a global technology leader with many firsts as an African-American woman, including the first African-American woman to receive a PhD in Electrical Engineering with a focus on Computer Engineering in the United States. This talk discusses the open doors, obstacles, and self-direction she experienced and encountered throughout her journey to current success. This includes mentors and others who assisted with her professional development, as well as those who sought her derailment.

Machine learning tools promise to help solve data curation problems. While the principles are well understood, the engineering details in configuring and deploying ML techniques are the biggest hurdle. In this talk, I discuss why leveraging data semantics and domain-specific knowledge is key in delivering the optimizations necessary for truly scalable ML curation solutions.

Knowledge graphs have been used to support a wide range of applications and enhance search results for multiple major search engines, such as Google and Bing. At Amazon we are building a Product Graph, an authoritative knowledge graph for all products in the world.

We study the problem of secure two-party computation of arithmetic circuits. This problem is motivated by privacy-preserving numerical computations, such as ones arising in the context of machine learning training and classification. Recent works on the problem have mainly focused on passively secure protocols, whose security holds against passive (``semi-honest'') parties but may completely break down in the presence of active (``malicious'') parties who can deviate from the protocol.

Providing a satisfactory quality of experience (QoE) to Internet users is crucial for content and service providers. When users get bad QoE from an application, such as the videos they are watching on a streaming provider keep freezing or the shopping Web site they are visiting takes a long time to load, they often spend less time on the application, return to it less frequently, or even worse they might switch to an alternative application, in all cases hurting the business financially.

We propose an architecture based on Convolutional Neural Networks (CNNs) for the detection of motion boundaries from two consecutive images of a video sequence. Existing learning-based approaches start with dense optical flow estimates, which are expensive to compute and often fail near motion boundaries, exactly where they are needed most. In contrast, we explore ways to detect motion boundaries without first computing optical flow. For efficiency, we hypothesize that motion boundaries occur at or near the edges of superpixels in an over-segmentation of the first image.

DNS enables everything else on the Internet -- both good and bad. By watching what bad guys do with their DNS configurations and offering them differentiated (that is to say, poor) service, defenders can re-level the playing field in our favor. In this keynote address, Internet pioneer and CEO of Farsight Security, Inc., Dr. Paul Vixie, CEO of Farsight Security, will explain how the bad guys are exploiting DNS to commit fraud and other cybercrime and specific techniques you can use i.e. DNSSEC, TSIG, RRL and RPZ to defend your organization.

One of the most pressing challenges in genomics is to reconstruct a long and contiguous DNA sequence from short DNA subsequences (contigs). Enabled by very recent developments in sequencing technology one can now obtain linkage information that is statistically correlated with the true contig ordering, so that many links are observed between neighboring pairs of contigs, while relatively few links are observed for non-neighboring pairs. 

Many datasets have a temporal dimension and contain a wealth of historical information. Temporal data tends to be noisy and often exhibits transient patterns in real life, e.g., stock market, temperature monitoring, etc. When using such data to make decisions, we often want to examine not only the current snapshot of the data but also its history. For example, given a result object of a snapshot query, we can ask for its durability, or intuitively, how long (or how often) it was valid in the past.

Faculty and graduate students are invited to a roundtable discussion with Gyana R. Parija, IBM’s Manager of Analytics & Optimization Research and Global Research Lead, Collaborative Cognition Research at IBM Research. At this roundtable, Gyana will discuss recent IBM research initiatives in AI, and then explore areas of mutual interest for possible collaboration. Please consider attending if your research interests include these topics.