Game theory:

• Markus Brede, The evolution of cooperation in structured populations
• Starting with a brief introduction to evolutionary game theory, I will discuss some ideas that can explain how cooperative behaviour can evolve and persist in populations of interacting agents. A focus of the overview will be on structured populations and so-called "network reciprocity" and I will present some of my own work that highlights the importance of various types of heterogeneity for the evolution of cooperation. I'll conclude by discussing some recent directions and advances in the field.

• Valerio Restocchi, Measuring market composition through the favourite-longshot bias
• The favourite-longshot bias is a recurring anomaly affecting time-limited markets, in which those events characterised by high (low) probability of realisation are, on average, underpriced (overpriced). I will talk about the model we developed for explaining this anomaly, showing how skewing prices and therefore causing the favourite-longshot bias is the best strategy for the market maker if the market is populated by heterogeneous agents. I will also show how this model can be used as a tool for getting new insights about time-limited markets, especially about market composition and efficiency.
• Laurie Carver, Regulation of systemic risk in financial networks (slides)
• The financial crisis of 2008 cost the UK, US and some EU governments several trillion dollars in public "bail out" subsidies. It was so bad and so broad because of "systemic risk" - the network of debt interlinking bank balance sheets allowed for the contagion of losses between them. Some attempts to tax banks' systemic riskiness have been proposed, to try to ensure they pay more of the costs of future bail outs. But which tax is best depends on how the network will react to it, and there are at present no good models for analysing that. I am developing one, based on multilateral bargaining games - with the principal twist being the network effect means the bargaining set is not convex, so standard solutions like the Nash solution should not be naively applied.

• Sofia Ceppi, Auction Mechanism for Mixed Ads (slides)
• The sponsored search auction currently used by search engines to decide which ads to display and how much charge advertisers is a variation of the Generalized Second Price (GSP) auction. Recent studies show that this type of auction is no longer suitable because of the rich space of ads types (e.g., images, rich media) and their combinations that are nowadays available. Truthful auctions are not affected by problems that make the GSP auction not appropriate anymore, and, thus, search engines should be motivated to use them. However, suddenly changing the auction used, moving from the GSP to a truthful one, can entail a significant loss in revenue for the search engine. In the talk, I will describe a hybrid auction mechanism to use during the transition from the GSP auction to a truthful one that overcomes the problem of the loss in revenue of the search engine. With this mechanism bidders can be truthful or not and are accordingly treated differently. In particular, the class of hybrid mechanisms gives incentives for non-truthful bidders to bid truthfully, while behaving as a non-truthful auction if no bidder is truthful.

• Tim Baarslag, Pandora's Problem - a technique from search theory to deal with uncertainty and cost (slides)

• In this talk, I will introduce a problem from search theory called Pandora Problem. The problem deals with optimizing reward while searching through boxes with uncertain prizes, which can be uncovered against a certain cost. Pandora's Problem not only has a wide range of possible applications, it also has a surprisingly elegant and optimal solution. I will provide some intuition behind the solution and then show an application of this technique in the form of a negotiating agent that can adapt itself to the user. In such a negotiation setting, it is essential for the agent to understand the user's preferences, without exposing them to elicitation fatigue. I discuss a new model, using Pandora's Rule, in which a negotiating agent may incrementally elicit the user's preference during the negotiation.  This yields an optimal elicitation strategy that decides, at every stage of the negotiation, how much additional user information to extract at a certain cost.

• Long Tran-Thanh, (Tutorial) Evolutionary Game Theory (EGT)
• EGT is a branch of game theory which focuses on the dynamics of strategy changes of the agents within complex and evolving systems. Originally introduced as a framework for modelling Darwinian competition that occur in biological systems, it has proven to be a good tool to explain the existence of cooperation and altruism, including their underlying dynamics, which was not able to be done within classical game theory. In this brief tutorial, I will provide a basic introduction to EGT, with a focus on the evolution of cooperation.

• Filippo Bistaffa, Coalition Formation in Multi-Agent Systems (slides)
• Coalition formation represents a powerful tool in cooperative game theory, used to model cooperations among agents in scenarios in which they cannot complete tasks by themselves. In this tutorial, I will provide a brief introduction on the main concepts regarding coalition formation and on the state of the art techniques that can be adopted to compute solutions. In particular, I will address the 3 main tasks involved in the coalition formation process:
• coalitional value calculation: defining a characteristic function which, given a coalition as an argument, provides its coalitional value;
• coalition structure generation: finding a partition of the set of agents (into disjoint coalitions) that maximises the sum of the values of the chosen coalitions;
• payment computation: finding the transfer or payment to each agent to ensure it is fairly rewarded for its contribution to its coalition.

• Maria Polukarov, (Tutorial) Strategic Voting and Strategic Candidacy in Multi-Agent Systems (slides)
• Multi-agent decision problems, in which independent agents have to agree on a joint plan of action or allocation of resources, are central to various applications. Often in such settings, agents' individual preferences over available alternatives may vary, and they may try to reconcile these differences by voting. Based on the fact that agents may have incentives to vote strategically and misreport their real preferences, much of the literature in computational social choice focuses on evaluating voting rules by their resistance to strategic behaviours and uses computational complexity as a barrier to them. In contrast, more recent works (counting from 2010) take another natural approach and analyse voting scenarios from a game-theoretic perspective, viewing strategic parties as players and examining possible stable outcomes of their interaction (i.e., equilibria). Finally, the candidates themselves may also have preferences about the outcome and try to affect it by strategically choosing whether to stand for election or not.
  
  This tutorial will begin by introducing the audience to basic notions of social choice and game theory, and will lead them to an understanding of strategic behaviours by voters and candidates, modelling such scenarios as voting/candidacy games, and analysing the existence of stable game outcomes (including those with predetermined properties), as well as their reachablity by natural iterative processes, such as best-response dynamics or its restricted variants. Convergence of such procedures is a highly desirable property of the game, since, from a system-wide perspective, it implies that a system has a deterministic stable state that can be reached by the agents without any centrallised control and/or communication.
  
  The tutorial will give an overview of the existing results and present a number of open problems in this---fairly new---line of research.


• Enrico Gerding, (Tutorial) Online Mechanism Design (slides)

• In traditional allocation mechanisms, including standard auctions and the VCG mechanism, it is assumed that agents are always present in the market or that all arrival and departure times are known in advance. However, in many realistic settings, this does not hold. Bidders arrive dynamically over time, and the mechanism needs to gradually allocate resources without perfect foresight of future arrivals. This tutorial will discuss how to design incentive compatible mechanisms in these settings.


• Dengji Zhao, Mechanism Design for Ridesharing (slides)
• This talk will present a novel market-based system for ridesharing, where commuters are matched based on their declared travel constraints, the number of available seats (which could be zero), and their costs. Based on this information, the system then designates commuters to be either drivers or riders, finds appropriate matches, and calculates rewards for drivers and payments for riders. We show that, for this system, the well-known Vickrey-Clarke-Groves (VCG) mechanism is incentive compatible (IC), individually rational (IR) and efficient (i.e., minimizing cost), but results in a very high deficit, thus requiring large subsidies. We therefore investigate alternative mechanisms. This talk will specifically give some details about our AAAI15 paper: Balanced Trade Reduction for Dual-Role Exchange Markets, which is motivated by McAfee's trade reduction (McAfee, 1992).


• Seb Stein, (Tutorial) Mechanism Design (slides)
• In many multi-agent systems, scarce resources have to be allocated to competing agents, who each have their own private valuation and constraints for using these resources. Examples include allocating computational processing capacity to cloud service consumers, scheduling the charging of electric vehicles or selling physical goods to multiple bidders. Mechanism design is a sub-field of game theory, which deals with exactly these types of settings and provides tools for designing interaction mechanisms that ensure certain desirable properties (such as maximising the social welfare or limiting the scope for strategic misreporting of private information). This talk will provide a basic introduction to the field of mechanism design, covering a number of widely-used mechanisms, including auctions and the well-known VCG mechanism.

• Mahesan Niranjan,  Excitement at the machine learning - biology interface (slides)
• I will introduce some interesting and challenging machine learning problems my team is working on, using  examples of problems in biology that trigger the search for new machine learning approaches. 


• Jose Alcala Orzaez, Health Monitoring of Elderly Residents using Disaggregated Smart Meter Data and Log Gaussian Cox Processes

• Monitoring the health of the elderly living independently in their own homes is a key issue in building sustainable healthcare models which support a country’s ageing population. Existing approaches have typically proposed remotely monitoring the behaviour of a household’s occupants through the use of additional sensors. However the costs and privacy concerns of such sensors have significantly limited their potential for widespread adoption. In contrast, in this talk we propose an approach which detects appliance usage from existing smart meter data, from which the unique daily routines of the household occupants are learned automatically via a log Gaussian Cox process. We evaluate our approach using two real-world data sets, and show it is able to detect over 80% of appliance uses while generating less than 20% false positives. Furthermore, our approach allows earlier interventions in households with a consistent routine and fewer false alarms in the remaining households, relative to a fixed-time intervention benchmark..


• Filippo Bisfatta, Discussion of AAAI 2015 papers
• I will report back from AAAI 2015, the Twenty-Ninth AAAI Conference on Artificial Intelligence. AAAI is one of the main Artificial Intelligence conferences, which covers lots of different topics which I think could be of general interest to this reading group. In particular, during this talk I will present the main topics covered by a number of papers, together with the solution approaches and the achieved results.


• Oliver Parson, NILMTK: An Open Source Toolkit for Non-intrusive Load Monitoring (Best demo award at BuildSys 2014)
• Non-intrusive load monitoring, or energy disaggregation, aims to separate household energy consumption data collected from a single point of measurement into appliance-level consumption data. I this talk, I will present the Non-intrusive Load Monitoring Toolkit (NILMTK); an open source toolkit designed specifically to enable the comparison of energy disaggregation algorithms in a reproducible manner. This work is the first research to compare multiple disaggregation approaches across multiple publicly available data sets. Our toolkit includes parsers for a range of existing data sets, a collection of preprocessing algorithms, a set of statistics for describing data sets, two reference benchmark disaggregation algorithms and a suite of accuracy metrics. I will also discuss the differences in publishing a paper based on a software toolkit in comparison with more traditional papers based on algorithms.