Artificial Intelligence

Analysis, modeling, and control of dynamic processes in networks

Friday, December 01, 2017, 11:00am - 12:00pm PDTiCal
11th floor large conference room
This event is open to the public.
AI Seminar - Interview talk
Victor Amelkin
Network processes have surrounded people for thousands of years. Heat diffusing through a complex material, traffic jams propagating through a road network, commodities spreading through a network of merchants, people passing rumors via word of mouth, collaboration patterns arising inside complex organizations are just a few examples. A special place is occupied by the processes in social and economic networks, as they are immediately connected to people's social and financial well-being. Most recently, with the emergence of massive-scale online social networks, and trading and collaboration platforms and tools, studying network processes has become feasible and important as never before. In this talk, I will summarize how my research contributes to the studies of network process along three directions---(i) model-based analysis of observed network processes, as well as network process' (ii) modeling, and (iii) control---focusing on the opinion formation and spread processes in large-scale online social networks. Some open problems will also be discussed.
Firstly, I will review a general framework for the analysis of observed network processes, and highlight the concept of the Social Network Distance---a distance measure that quantifies the amount of change happening between two time points in a large evolving social network with respect to an opinion dynamics model defining the expected opinion spread scenarios. This work---relying on specialized shortest path and combinatorial network flow algorithms---resulted in a pseudo-linear-time method that effectively detects anomalies---such as viral marketing campaigns---in the opinion formation process.
Secondly, I will talk about a class of non-linear polar opinion formation models that capture the dynamic dependency of human resilience to persuasion upon the held beliefs. For example, a person shifting towards the Republican ideology may be harder to persuade to invert her or his political stance; alternatively, in a society with strong social norms, conservative opinions may be more resilient, while extreme opinions are volatile. I will review the theoretical core of this work---comprised of non-smooth analysis of dynamical systems---and touch upon a few qualitative findings about the opinion formation process.
Thirdly, I will review a novel problem of disabling attempts of malicious intervention into the opinion formation process in large-scale social networks, where the attacker tries to manipulate the "average opinion" by influencing some users, while our goal is to eliminate the effect of such attacks by strategically recommending a small number of new social ties to the network's members. The theoretical core of this work is comprised of perturbation analysis of stationary distributions of Markov chains, and efficient estimation of mean first passage times in large chains, while the practical outcome is a pseudo-linear-time heuristic that effectively solves the defined above large-scale social intervention problem.
Victor Amelkin is a PhD Candidate in the Department of Computer Science at the University of California, Santa Barbara (UCSB), advised by Ambuj K. Singh. He works on analysis, modeling, and control of dynamic processes in social and economic networks, at the intersection of domain-specific areas, such as social psychology and behavioral economics, as well as the fundamental fields of combinatorial algorithm design, linear algebra, and dynamical systems theory. Prior to joining UCSB, Victor had spent several years as a software engineer in industry. He obtained his Master's degree in Applied Mathematics and Computer Science from Tula State University, Russia.
« Return to Events