Signal Processing I

B1L-B: Signal Processing I

Session Type: Lecture
Session Code: B1L-B
Location: Room 2
Date & Time: Thursday March 23, 2023 (09:00 - 10:00)
Chair: Tugba Erpek
Track: 3
Paper No. Paper NameAuthorsAbstract
3079Distributed Detection with Unreliable Reporting Channels: Quantize or Not Quantize?Lei Cao, Ramanarayanan ViswanathanIt is known that reporting the raw data observed at sensors to the fusion center completely and without any channel distortion always performs better than reporting the quantized data. However, when channel noise exists as it is always in practice, we show whether it is optimal to send the unquantized raw data or quantized data is a very interesting question that requires more research efforts. We compare these two schemes under the constraints of using the same transmission power and the bandwidth where analog modulation is assumed per channel use. The detection performance with one sensor and two sensors is presented in terms of the Bayes error. The performance of using asymptotically many sensors is compared with the Chernoff information. What presented in the paper could open up a new perspective in the design of the optimal distributed detection systems.
3086Outlier Detection for Generative Models with Performance GuaranteesJingchao Gao, Jirong Yi, Weiyu XuWe consider the problem of recovering signals using deep generative models, from measurements contaminated with sparse outliers. We propose an optimization based outlier detection approach for reconstructing the ground truth signals modeled by generative models under sparse outliers. We further establish theoretical recovery guarantees for our proposed reconstruction approach under outliers. Our results are applicable to a broad class of generative neural networks with an arbitrary number of layers. The experimental results show that the signals can be successfully reconstructed under outliers using our approach.
3040Optimal Decision Making with Rational Inattention Using Noisy DataYuan Zhao{3}, Atah Abdi{2}, Mark Dean{1}, Ali Abdi{3}Rational inattention of decision makers to costly data and information and resources affects their optimal decision making strategies. The theory of rational inattention has found applications in several areas such as economics, finance and psychology. In this paper, we study scenarios where the available data is noisy. The noise may have been generated because of inaccuracies or errors in data collection methods, or the data may have been intentionally distorted to protect private or secret information. Here we introduce a formulation for rationally inattentive decision making when the data is noisy, and derive its optimal decision making strategy. Using a stock trading problem as an example, we demonstrate that as the noise level in the data increases, probability of correct decision decreases. This results in less payoff for the decision maker, when using noisy data. We also show how the noise level and information cost parameters can be estimated using the developed formulation. The results are useful for developing decision making strategies, when using noisy data.