Experience
Attacking Scene Flow in 3D Point Clouds (to be submitted)
Authors: Haniyeh Ehsani Oskouie, Mohammad-Shahram Moin, and Shohreh Kasaei
Abstract: Deep neural networks have made significant advancements in accurately estimating scene flow using point clouds, which is vital for safety-critical applications like autonomous vehicles. The robustness of these techniques, however, remains a concern, particularly in the face of adversarial attacks that have been proven to deceive state-of-the-art deep neural networks in many domains. Surprisingly, the robustness of scene flow networks against such attacks has not been thoroughly investigated. In this paper, we aim to bridge this gap by introducing adversarial white-box attacks specifically tailored for scene flow networks. Our findings demonstrate that these attacks can indeed compromise the performance of scene flow networks, highlighting the need to address their vulnerabilities in this domain. Our study also reveals the significant impact that attacks targeting point clouds in only one dimension have on the overall results.
Publications:
Interpretation of Neural Networks is Susceptible to Universal Adversarial Perturbations (ICASSP 2023)
Authors: Haniyeh Ehsani Oskouie, and Farzan Farnia
Abstract: Interpreting neural network classifiers using gradient-based saliency maps has been extensively studied in the deep learning literature. While the existing algorithms manage to achieve satisfactory performance in application to standard image recognition datasets, recent works demonstrate the vulnerability of widely-used gradient-based interpretation schemes to norm-bounded perturbations adversarially designed for every individual input sample. However, such adversarial perturbations are commonly designed using the knowledge of an input sample, and hence perform sub-optimally in application to an unknown or constantly changing data point. In this paper, we show the existence of a Universal Perturbation for Interpretation (UPI) for standard image datasets, which can alter a gradient-based feature map of neural networks over a significant fraction of test samples. To design such a UPI, we propose a gradient-based optimization method as well as a principal component analysis (PCA)-based approach to compute a UPI which can effectively alter a neural network's gradient-based interpretation on different samples. We support the proposed UPI approaches by presenting several numerical results of their successful applications to standard image datasets.
Collaboration on a paper: The Psychological Effects of the Home Environment during Self-Quarantine: a Web-based Cross-Sectional Survey in Iran (submitted to IJAUP 2023)
Authors: Jamal-E-Din Mahdi Nejad, Hamidreza Azemati, Seyede Fereshteh Ehsani Oskouei, and Zinat Aminifar
Abstract: During the COVID-19 outbreak in Iran, self-quarantine was a measure to slow the spread of this infection. We conducted this cross-sectional study to explore the psychological effects of the home environment while people had to stay at home for a long time. For the survey, 536 individuals took part. Collecting data was via an online questionnaire including three sections: (1) Demographic characteristics and general information; (2) Home environment features and (3) Negative psychological experiences (NPE) considered as (a) feeling of sadness and depression; (b) feeling of stress and anxiety; and, (c) experiencing domestic violence during quarantine. For data analysis, first, some descriptive information about the participants was presented; then, we used a logistic regression model, one of the classification algorithms in machine learning methods to investigate the association of home environment features and NPE during self-quarantine. The results indicate the home environment affects NPE differently among men and women. Generally, the individuals who were more satisfied with their house performance during quarantine, and people considered the light quality of their house as appropriate; besides, residents with less noise disturbance issues had a better mood during this period. Conversely, failure in the possibility of indoor exercising and the feeling of being in a crowded house increased the level of NPE.
I helped with implementing the methods in Python language (using Scikit-learn and Pandas).
Collaboration on a paper: Effects of Residential Units’ Characteristics on Mental Health of Residents during Quarantine in Iran (to be submitted)
Authors: Jamal-E-Din Mahdi Nejad, and Fereshteh Ehsani Oskouei
Description: This paper is targeting the research area of Architecture Design. It uses ML approaches to show what residential unit features have the most influence in making residents uncomfortable during quarantine.
I helped with implementing the methods in Python language (using Scikit-learn and Pandas).
Research Experience:
Research Assistant at the University of California, Los Angeles
September 2023 - Present
Supervisor: Prof. Majid Sarrafzadeh
Description: In this research, we want to experiment with how cross-model neuronal correlation can be used to indicate model generalizability. We intend to answer these questions:
Is there any meaningful relationship between correlation and performance across models (i.e., can we establish a relationship between model correlation and associated performance) Is this dependent on the same underlying dataset? Is this dependent on the same feature set (but possibly a different subset of data)? Is this dependent on uniform overall model construction, or is correspondence between one layer sufficient?
Junior Research Assistant at the Chinese University of Hong Kong & Imperial College London
July 2023 - September 2023
Supervisor: Prof. Farzan Farnia and Prof. Seyed Mohsen Moosavi-Dezfooli
Description: In this research, we aimed to discover a universal adversarial perturbation for the interpretation (UPI) of neural networks while ensuring that the perturbations do not alter the classification decisions of the models.
Undergraduate Research Assistant at the University of Birmingham
March 2022 - April 2023
Supervisor: Prof. Mohan Sridharan
Description: Our goal in this research was optimizing an ad hoc teamwork (AHT) problem in the FortAttack domain. An important observation in the ad hoc teamwork (AHT) domain is that deep neural networks have not demonstrated a significant advantage over simple machine learning approaches. For this reason, we employed Multinomial Logistic Regression with the STEW loss function. This approach allowed us to effectively model and improve the performance of the AHT system in the FortAttack domain.
Bachelor Thesis
October 2021 - February 2023
Supervisor: Prof. Shohreh Kasaei
Description: In this project, I studied the effects of attention layers and deformable convolutions on optical flow estimation. I explored different loss functions and investigated the robustness of scene flow estimation networks against adversarial attacks.
Summer Intern at the Chinese University of Hong Kong
July 2022 - October 2022
Supervisor: Prof. Farzan Farnia
Description: In this work, we focused on finding a universal adversarial perturbation for the interpretation (UPI) of neural networks. To design such a UPI, we propose a gradient-based optimization method as well as a principal component analysis (PCA)-based approach to compute a UPI which can effectively alter a neural network's gradient-based interpretation on different samples.
Undergraduate Research Assistant at Sharif University of Technology
March 2021 - December 2021
Supervisor: Hamid R. Rabiee
Description: In this project, we focused on cancer detection by employing representation learning and semantic segmentation techniques on CennaLab's dataset. we explored different models, including ResNet-50 for supervised learning and SwAV for unsupervised learning, to improve the accuracy of the detection system. For this purpose, we leveraged the strengths of SwAV by integrating it as the encoder component of the U-Net architecture, aiming to enhance the efficiency and performance of the segmentation task. By combining representation learning and semantic segmentation approaches, we aimed to develop an effective and efficient cancer detection system that can contribute to advancements in medical imaging and diagnosis.
