I am Mahed Abroshan currently a postdoc at Alan Turing Institute. My research topic is trustworthy machine learning (concepts like interpretability, fairness, and privacy). In particular, I worked on applying ML in healthcare (where trustworthiness is a must). I did my PhD at Cambridge working on a problem in information theory (insertion/deletion channels). In general I am passionate about problem solving and research on fundamental problems with real-life impact, for which my background in math is very useful (e.g. I was in math competition business when I was younger).

Please download my CV here (updated in January 2022)

Research

I recently joined FAIR project at Turing working on responsible AI and in particular on synthetic data generation.

Before that, I was part of the Learning Machine project working on the deployment of machine learning in high-stake applications like healthcare and the criminal justice system. We were particularly focused on monitoring the model for use for a long period (looking for different sorts of drifts). I was working with a group of research engineers and my role was helping with the design and implementation of the model. We have worked on a few datasets including the SEER dataset (for cancer research) and ADNI dataset (Alzheimer’s disease). We have also collaborated with clinicians in our projects. I have worked on several problems related to the learning machine project including interpretability, fairness and privacy, and semi-supervised learning with covariate shift (see publications).

Before joining Turing I was shortly a postdoc in van der Schaar Lab at Cambridge (department of applied mathematics and theoretical physics) where I worked on treatment recommendations with missing values and also a short project on the application of ML in cystic fibrosis.

My research topic during my PhD and Master’s studies was information theory. In my PhD, I worked on the problem of the insertion-deletion channel and file synchronization with application in DNA storage systems and file sharing platforms. My Master’s thesis was on zero error coordination.

Publication

Machine Learning

• (Under review) M. Abroshan, S. Mishra, M.M. Khalili, “A Memetic algorithm for Interpreting Black- boxes Using Primitive Functions”

• (Under review) M.M. Khalili, X. Zhang, M. Abroshan, I. Vakilinia, “Non-convex Optimization for Learning a Fair Predictor under Equalized Loss Fairness Constraint”

• (Accepted) M. Abroshan*, G. Aminian*, M.M. Khalili, L. Toni, M. Rodrigues, “An Information Theory Approach to Semi-supervised Learning under Covariate-shift”, AISTAT 2022 (* equal contribution)

• M. Abroshan, K. Yip, C. Tekin, M. van der Schaar, “Conservative Policy Construction Using Variational Autoencoders for Logged Data with Missing Values” IEEE Transactions on Neural Networks and Learning Systems (2022)

• M.M. Khalili, X. Zhang, M. Abroshan, “Fair Sequential Selection Using Supervised Learning Models” Advances in Neural Information Processing Systems, Neurips 2021

• M.M. Khalili, X. Zhang, M. Abroshan, S. Sojoudi, “Improving Fairness and Privacy in Selection Problems” Association for the Advancement of Artificial Intelligence AAAI 2021

• M. Abroshan, A. M. Alaa, O. Rayner, M. van der Schaar, “Opportunities for machine learning to transform care for people with cystic fibrosis ” (editorial) Journal of Cystic Fibrosis (2020)

Information Theory

• M. Abroshan, R. Venkataramanan, A. Guillén i Fàbregas, “Multilayer Codes for Synchronization from Deletions and Insertions” IEEE Transactions on Information Theory (2020)

• M. Abroshan, R. Venkataramanan, A. Guillén i Fàbregas, “Coding for Segmented Edit Channels” IEEE Transactions on Information Theory (2018)

• M. Abroshan, R. Venkataramanan, L. Dolecek, A. Guillén i Fàbregas, “Coding for Deletion Channels with Multiple Traces” IEEE International Symposium on Information Theory 2019

• M. Abroshan, R. Venkataramanan, A. Guillén i Fàbregas, “Efficient Systematic Encoding of Non- Binary VT codes” IEEE International Symposium on Information Theory 2018

• M. Abroshan, R. Venkataramanan, A. Guillén i Fàbregas, “Codes for Channels with Segmented Edit” IEEE International Symposium on Information Theory 2017

• M. Abroshan, R. Venkataramanan, A. Guillén i Fàbregas, “Multilayer Codes for Synchronization from Deletions” IEEE Information Theory Workshop 2017, http://arxiv.org/abs/1705.06670

• M. Abroshan, A. A. Gohari, S. Jaggi, ‘‘Zero Error Coordination” IEEE Information Theory Workshop 2015, https://arxiv.org/pdf/1505.01110v1.pdf