Yu-Han Chang

Mobilized ad-hoc networks: a reinforcement learning approach

With the cost of wireless networking and computational power rapidly dropping, mobile ad-hoc networks will soon become an important part of our societys computing structures. While there is a great deal of research from the networking community regarding the routing of information over such networks, most of these techniques lack automatic adaptivity. The size and complexity of these networks demand that we apply the principles of autonomic computing to this problem. Reinforcement learning methods can be used to control both packet routing decisions and node mobility, dramatically improving the connectivity of the network. We present two applications of reinforcement learning methods to the mobilized ad-hoc networking domain and demonstrate some promising empirical results under a variety of different scenarios in which the mobile nodes in our ad-hoc network are embedded with these adaptive routing policies and learned movement policies.

On the utility of network coding in dynamic environments

Many wireless applications, such as ad-hoc networks and sensor networks, require decentralized operation in dynamically varying environments. We consider a distributed randomized network coding approach that enables efficient decentralized operation of multi-source multicast networks. We show that this approach provides substantial benefits over traditional routing methods in dynamically varying environments. We present a set of empirical trials measuring the performance of network coding versus an approximate online Steiner tree routing approach when connections vary dynamically. The results show that network coding achieves superior performance in a significant fraction of our randomly generated network examples. Such dynamic settings represent a substantially broader class of networking problems than previously recognized for which network coding shows promise of significant practical benefits compared to routing.

Network monitoring in multicast networks using network coding

In this paper we show how information contained in robust network codes can be used for passive inference of possible locations of link failures or losses in a network. For distributed randomized network coding, we bound the probability of being able to distinguish among a given set of failure events, and give some experimental results for one and two link failures in randomly generated networks. We also bound the required field size and complexity for designing a robust network code that distinguishes among a given set of failure events

TESLA: an extended study of an energy-saving agent that leverages schedule flexibility

This paper presents transformative energy-saving schedule-leveraging agent (TESLA), an agent for optimizing energy usage in commercial buildings. TESLA’s key insight is that adding flexibility to event/meeting schedules can lead to significant energy savings. This paper provides four key contributions: (i) online scheduling algorithms, which are at the heart of TESLA, to solve a stochastic mixed integer linear program for energy-efficient scheduling of incrementally/dynamically arriving meetings and events; (ii) an algorithm to effectively identify key meetings that lead to significant energy savings by adjusting their flexibility; (iii) an extensive analysis on energy savings achieved by TESLA; and (iv) surveys of real users which indicate that TESLA’s assumptions of user flexibility hold in practice. TESLA was evaluated on data gathered from over 110,000 meetings held at nine campus buildings during an 8-month period in 2011–2012 at the University of Southern California and Singapore Management University. These results and analysis show that, compared to the current systems, TESLA can substantially reduce overall energy consumption.

Mobility Patterns and User Dynamics in Racially Segregated Geographies of US Cities

In this paper we try to understand how racial segregation of the geographic spaces of three major US cities (New York, Los Angeles and Chicago) affect the mobility patterns of people living in them. Collecting over 75 million geo-tagged tweets from these cities during a period of one year beginning October 2012 we identified home locations for over 30,000 distinct users, and prepared models of travel patterns for each of them. Dividing the cities’ geographic boundary into census tracts and grouping them according to racial segregation information we try to understand how the mobility of users living within an area of a particular predominant race correlate to those living in areas of similar race, and to those of a different race. While these cities still remain to be vastly segregated in the 2010 census data, we observe a compelling amount of deviation in travel patterns when compared to artificially generated ideal mobility. A common trend for all races is to visit areas populated by similar race more often. Also, blacks, Asians and Hispanics tend to travel less often to predominantly white census tracts, and similarly predominantly black tracts are less visited by other races.

The Social Ultimatum Game

The Ultimatum Game is a key exemplar that shows how human play often deviates from “rational” strategies suggested by game-theoretic analysis. One explanation is that humans cannot put aside the assumption of being in a multi-player multi-round environment that they are accustomed to in the real world. In this paper, we introduce the Social Ultimatum Game, where players can choose their partner among a society of agents, and engage in repeated interactions of the Ultimatum Game. We provide theoretical results that show the equilibrium strategies under rational actor models for the Social Ultimatum Game, which predict “unfair” offers as the stable solution. We develop mathematical models of human play that include “irrational” concepts such as fairness, reciprocity, and adaptation to social norms. We investigate the stability of maintaining a society of “fair” agents under these conditions. Finally, we discuss experimental data from initial human trials of the Social Ultimatum Game.

Hedged learning: regret-minimization with learning experts

In non-cooperative multi-agent situations, there cannot exist a globally optimal, yet opponent-independent learning algorithm. Regret-minimization over a set of strategies optimized for potential opponent models is proposed as a good framework for deciding how to behave in such situations. Using longer playing horizons and experts that learn as they play, the regret-minimization framework can be extended to overcome several shortcomings of earlier approaches to the problem of multi-agent learning.

Following Human Mobility Using Tweets

The availability of location-based agent data is growing rapidly, enabling new research into the behavior patterns of such agents in space and time. Previously, such analysis was limited to either small experiments with GPS-equipped agents, or proprietary datasets of human cell phone users that cannot be disseminated across the academic community for followup studies. In this paper, we study the movement patterns of Twitter users in London, Los Angeles, and Tokyo. We cluster these agents by their movement patterns across space and time. We also show that it is possible to infer part of the underlying transportation net- work from Tweets alone, and uncover interesting differences between the behaviors exhibited by users across these three cities.

No regrets about no-regret

No-regret is described as one framework that game theorists and computer scientists have converged upon for designing and evaluating multi-agent learning algorithms. However, Shoham, Powers, and Grenager also point out that the framework has serious deficiencies, such as behaving sub-optimally against certain reactive opponents. But all is not lost. With some simple modifications, regret-minimizing algorithms can perform in many of the ways we wish multi-agent learning algorithms to perform, providing safety and adaptability against reactive opponents. We argue that the research community should have no regrets about no-regret methods.

Gang Networks, Neighborhoods and Holidays: Spatiotemporal Patterns in Social Media

Social media generated by location-services-enabled cellular devices produce enormous amounts of location-based content. Spatiotemporal analysis of such data facilitate new ways of modeling human behavior and mobility patterns. In this paper, we use over 10 millions geo-tagged tweets from the city of Los Angeles as observations of human movement and apply them to understand the relationships of geographical regions, neighborhoods and gang territories. Using a graph based-representation of street gang territories as vertices and interactions between them as edges, we train a machine learning classifier to tell apart rival and non-rival links. We correctly identify 89% of the true rivalry network, which beats a standard baseline by about 30%. Looking at larger neighborhoods, we were able to show that distance traveled from home follows a power-law distribution, and the direction of displacement, i.e., the distribution of movement direction, can be used as a profile to identify physical (or geographic) barriers when it is not uniform. Finally, considering the temporal dimension of tweets, we detect events taking place around the city by identifying irregularities in tweeting patterns.