ICASSP'20

I’m happy to announce that our paper What Did Your Adversary Believe? Optimal Filtering and Smoothing in Counter-Adversarial Autonomous Systems has been accepted for presentation in a lecture session at the 45th International Conference on Acoustics, Speech, and Signal Processing (ICASSP). This year, the conference will be held 4-8 May, in Barcelona, Spain.

The post below contains an outline of the problem we consider in the paper.

The work has been developed together with Inês Lourenço, Vikram Krishnamurthy, Cristian Rojas and Bo Wahlberg. We are very excited about this line of research and are currently working on a number of extensions – stay tuned (and if you have any comments, let us know)!

What Did Your Adversary Believe?

A preprint of our latest work What Did Your Adversary Believe? Optimal Filtering and Smoothing in Counter-Adversarial Autonomous Systems is now available on arXiv.

An adversary deploys an autonomous filtering and control system that:

1. measures our current state via a noisy sensor,
2. computes a posterior estimate (belief) and
3. takes an action that we can observe.

Based on such observed actions and our knowledge of our state sequence, we aim to estimate the adversary’s past and current beliefs – this forms a foundation for predicting, and counteracting against, future actions.

Have a look, and feel free to send me any comments you may have by email!

ERNSI 2019

This week, I attended the annual workshop of the European Research Network on System Identification (ERNSI) in Maastricht, the Netherlands – very interesting presentations and discussions! Me and Inês Lourenço presented a poster related to our recent paper on estimating private beliefs from observed decisions.

Visiting Cornell University

I’m currently visiting Prof. Vikram Krishnamurthy at Cornell University in Ithaca, New York. The purpose of the visit is to continue our line of research on inverse filtering problems that was initiated with our NIPS 2017 paper and that has continued in the papers [1, 2].

In these problems, the standard filtering problem

Given observations from a noisy sensor along with a system model, compute estimates of the system’s state.

is flipped to

Given estimates (or actions based on these) of the state, determine how accurate the sensors and system model are.

Why are such questions of importance? Inverse filtering problems are partly motivated by the design of counter-autonomous systems: given measurements of the actions of a sophisticated autonomous adversary, how can our counter-autonomous system estimate the underlying belief of the adversary, predict future actions and, therefore, guard against these actions? Answers to these questions have potential applications in a vast range of fields: (cyber-)security, finance, social networks, to name a few – for more, see this recent paper that discusses a number of motivational real-world examples and proposes a Bayesian framework to inverse filtering.

Paper accepted for Measurement

Our paper A Framework for High-Resolution Frequency Response Measurement and Parameter Estimation in Microscale Impedance Applications has been accepted for publication in the Measurement journal. The work has been developed in collaboration with Roberto Ramírez-Chavarría and Matias Müller Riquelme.

In the paper, we propose a framework for spectral measurement and parameter estimation applied to electrical impedance spectroscopy (EIS) – an important tool for characterizing the electrical behavior of matter. The results have potential applications in biosensor systems (e.g., to measure bacterial concentration and to detect dangerous pathogens) as well as human health monitoring.

CDC'19

I’m glad to announce that our L-CSS paper Estimating Private Beliefs of Bayesian Agents Based on Observed Decisions has also been accepted for presentation at the 58th IEEE Conference on Decision and Control (CDC). The conference will this year be held 11-13 December in Nice, France.

In the paper, we consider sequential stochastic decision problems in which, at each time instant, an agent optimizes its local utility by solving a stochastic program and, subsequently, announces its decision to the world. Given this action, we study the problem of estimating the agent’s private belief – that is, its posterior distribution over the set of states of nature based on its private observations.

In other words: how can one estimate how agents perceive the world based on their decisions and actions? This is a natural continuation of our previous works on inverse filtering [1, 2].

L-CSS Accepted

I am happy to announce that our paper Estimating Private Beliefs of Bayesian Agents Based on Observed Decisions has been accepted for publication in the IEEE Control Systems Letters (L-CSS). The work has been developed in collaboration with Inês Lourenço.

In this paper, we investigate how one can estimate how agents perceive the world based on their decisions and actions. This is a natural continuation of our previous works on inverse filtering [1, 2]. We believe that our results have applications in, for example, social learning.

NVIDIA Course

I completed the course Fundamentals of Deep Learning for Computer Vision by the NVIDIA Deep Learning Institute. Very interesting to get some hands-on experience with these exciting new techniques. I’m looking forward to applying the concepts and learning more!

AC → DCS

The Department of Automatic Control is now the Division of Decision and Control Systems (DCS):

The Division of Decision and Control Systems conducts research, education and industry/society interplay within modeling, identification, control, learning and optimization of dynamical systems.

More information is available here.

Dept. of Automatic Control at NeurIPS

The Department of Automatic Control and Prof. Alexandre Proutiere (who I’m teaching EL2805 Reinforcement Learning together with) are currently featured in the KTH news:

Today marks the start of Neural Information Processing Systems (NeurIPS), the world’s largest and most prestigious conference in Artificial Intelligence (AI) and Machine Learning (ML), an arena to which merely the best are invited to present their research. […]

The Department of Automatic Control has published papers at NeurIPS consistently for six years, an achievement that is unique in Sweden. KTH in general did well, setting a record with five accepted papers. Three of those papers come from the Department of Automatic Control.

Last year, our department had two papers: mine and Stefan Magureanu’s paper.