Research Seminar Winter 2021
- Post by: Bahareh Arghavani
- November 29, 2022
- Comments off
| DATE | TIME | TITLE | PRESENTER | LINK |
| 29-Oct | 12pm-1pm EST | Adversarial Attacks on Deep Algorithmic Trading Policies | Nancirose Piazza | YouTube |
| 5-Nov | 12pm-1pm EST | ADVERSARIAL MANIPULATION OF EEG-BASED BCI | Upadhayay, Bibek | YouTube |
| 12-Nov | 12pm-1pm EST | Fault Detection and Prognosis in Medical Devices | Binesh Kumar | YouTube |
| 19-Nov | 12pm-1pm EST | Fairness in Fake News Detection Machine Learning Model | Yueyang Qin, Harsha Choudary | YouTube |
Adversarial Attacks on Deep Algorithmic Trading Policies
Abstract
Deep Reinforcement Learning (DRL) has become an appealing solution to algorithmic trading such as high-frequency trading of stocks and cryptocurrencies. However, DRL policies are shown to be susceptible to adversarial attacks. It follows that algorithmic trading DRL agents may also be compromised by such adversarial techniques, leading to policy manipulation. In this paper, we develop a threat model for deep trading policies and propose two active attack techniques for manipulating the performance of such policies at test-time. Additionally, we explore the exploitation of a passive attack based on adversarial policy imitation. Furthermore, we demonstrate the effectiveness of the proposed attacks against the benchmark and real-world DQN trading agents.
-Nancirose Piazza
ADVERSARIAL MANIPULATION OF EEG-BASED BCI
Abstract
Humanity is becoming increasingly connected to technology. The advent of the smartphone has accelerated the availability and reliance on human-computer interactions. Based on this trajectory the next logical step in the evolution of technology interaction will be through a Brain-Computer Interface (BCI). As BCI-enabled applications and devices increase in commercial viability, the threat of abuse also increases. Although there are many options for providing a hardware interface between the brain and technology the electroencephalogram (EEG) appears to be the most viable for mainstream adoption. Today the EEG provides a cost-effective mobile entry point for neurological research and integration with many commercial products. Therefore, our team is exploring attack vectors that may be approached by malicious actors in the future. Our research paper will explore the plausibility of a training time attack to induce a denial of service against a BCI-enabled interface. We leveraged the human brain as a proxy for our attack by using error-related potentials to induce perturbations. These minute perturbations influenced the machine learning model’s ability to classify the intentions of the user. This placed the ability to control the application in the hands of a potential malicious actor.
-Bibek Upadhayay
Fault Detection and Prognosis in Medical Devices
Abstract
The medical device industry is leaping forward by relying on electronics to improve life-saving medical devices’ safety and performance. The complex electronics, including microchips and FPGAs, run powerful software that helped further improve usability. However, the complex electro-mechanical systems introduced a new set of failure modes that are often difficult to identify and mitigate through traditional test protocols. The evolution of technology in connectivity and data collection paired with sensors opens the door for preventive and predictive maintenance to mitigate failure in critical devices. The predictive and preventive maintenance strategy uses fault detection techniques leveraging data, signal, process, or knowledge-based methods. These techniques detect and prevent faults that otherwise would result in a failure causing a safety issue or degraded performance. Through our independent study, we will be surveying the state of the art Fault Detection and Prediction algorithms and perform a feasibility study to understand the applicability to medical devices.
-Binesh Kumar