Study Group, University of Oxford, UK
In this talk, I began with maximum-likelihood–based parameter estimation for generalized linear models (GLMs) and discussed the robustness aspects of this approach. I then introduced α-divergence and showed how the estimation method can be extended using this robust measure. Finally, classical detectors were extended by incorporating this new estimation framework.
Analysis Group, University of Oxford, UK
Machine unlearning aims to give individuals data autonomy by removing their data’s influence from trained models. In this talk, I showed that these unlearning updates can paradoxically enable high-accuracy reconstruction attacks that recover the deleted data, especially for very simple models like linear regression. I also showed the generalization of these attacks across losses, architectures, and datasets, demonstrating that data deletion can introduce serious privacy risks even in settings previously considered safe.
Study Group, University of Oxford, UK
In this talk, I introduced Active Inference Theory, a mathematical framework that describes how self-organizing systems learn from and interact with their environment. I explained the concept of surprise functions, how they are typically defined, and why they cannot be minimized directly. Instead, I discussed how Variational Free Energy—an upper bound on surprise—can be minimized as a practical alternative. I also covered Expected Free Energy and demonstrated, through examples, how planning can be understood within the Active Inference framework.
Study Group, University of Oxford, UK
In this talk, I explained XAI taxonomy and talked about two popular techniques, i.e., Shapley values and Grad-CAM which have been widely used in the literature.
Study Group, University of Oxford, UK
In this talk, I explained the main idea behind the certification techniques in deep learning models (either classiffiers or regression models) against adversarial attacks. I also showed some synthetic and real world examples of the main theoritical results in the context of autonamous visual positioning systems.
DLSOD2022 Workshop, ACCV2022, Macau SAR, China
As organizers of this workshop, we discussed challenges in detection and localization of small objects in images and videos. In addition to that, A/Prof. Wanli Ouyang from The University of Sydney and Dr. Emre Akbas from Middle East Technical University (METU) presented their recent research in the field of object detection, with emphasis on small objects. (Link)
Poster Presentation, ICASSP2020, Barcelona, Spain
In this talk, we present an extension of generalized likelihood ratio test for the cases where observes signal is contaimaned with heavy impulsive noise. (video)
Poster Presentation, ICASSP2020, Barcelona, Spain
In this talk, we present an extension of classical adaptive mathched filter for the cases where the secondary set of data is not target-free and the covariance should be estimated differently to not include target signals. (video)
The University of Melbourne, Department of Electrical and Electronic Engineering, Melbourne, Australia
In this talk, I discussed the challenges of maximum likelihood-based estimators in facing contaminations and outliers. Several metrics were introduced for performance evaluation of a robust estimation method and finally, the well-known robust estimators such as the Huber estimator, Hample estimator, etc. were analyzed with respect to these metrics.