SOFWERX’s Post

Last week we presented at #IEEE #MILCOM2023 in Boston a new method for preventing eavesdroppers from learning the type of wireless communication signals. We assume eavesdroppers that deploy a DNN-based classifier for this task. The method is shown to be robust even to DNNs with adversarial training. You can find the paper here https://lnkd.in/dxEVSFSx

Please check out the BMVA meeting on Trustworthy Multimodal Learning with Foundation Models: https://lnkd.in/e3RkXDDM

🌟Exciting News!🌟 Hello Connections, I'm thrilled to announce that my research has been published in the 2024 5th International Conference on Innovative Trends in Information Technology (ICITIIT) by IEEE, held in Kottayam, India. The paper is titled "Efficient Light-weight Deep Learning Models for Drowsiness Detection." 🗓️ Conference Date: 15-16 March 2024 I am incredibly grateful to my fellow researchers Pranshul Hatwar and Gaurav Sahu for their collaboration and dedication throughout this research journey. A special thanks to our mentors, Miss Anjali Rajak Ma'am and especially Dr. Rakesh Tripathi Sir, for their guidance and support in making this publication possible. Here's a brief overview of our research: The increasing prevalence of global accidents highlights the crucial role of driver alertness in road safety. Our study focuses on behavioral measurements with minimal hardware requirements. We propose a drowsiness detection architecture using the UTA-RLDD dataset and MobileNet V2 as the base model for classifying drowsy and non-drowsy driver images. Compared to state-of-the-art models like Basic CNN, Xception, EfficientNetV2S, ResNet, and PeleeNet, our model achieves a remarkable 94.03% average accuracy with a training time of 60.6 seconds and a model size of 8.95 MB. This lightweight architecture can be efficiently used on mobile devices, reducing compute resources significantly. 🚀 Looking forward to more exciting research and innovations in the future! 🚀 You can follow up the link and read the research https://lnkd.in/dCGMAMSg. #Research #IEEE #ICITIIT2024 #DrowsinessDetection #DeepLearning #Innovation #RoadSafety #MobileNetV2

📣 We're excited to spotlight some of the most impactful articles published in the IEEE Journal on Emerging and Selected Topics in Circuits and Systems (JETCAS) from 2023. Whether you're a seasoned researcher or just starting, JETCAS offers valuable insights into the future of circuits and systems. 🔗 View the full list here: https://loom.ly/7CbAAU4 #JETCAS #CircuitsandSystems

Come to our INTERNATIONAL CHINESE STATISTICAL ASSOCIATION booth! Many #networking activities are in store throughout #JSM2024 through the #ICSA. Grace (Ying) Li, Zhezhen Jin, Jennifer Gulley, Kelly H. Zou, PhD, PStat, FASA, etc. #portland #oregon #JSM #statistician #statisticians #datascience #datasciences #datascientist #datascientisy AI4Purpose Inc. #AI4Purpose

Hi network, I've got some good news! 🎉 Recently, I’ve been awarded first place in the #SGSMA2024 Ph.D. Dissertation Competition! This recognition means a lot to me, and I'm looking forward to presenting my work at SGSMA 2024 next week. A big thank you to the committee and my supervisor Dr. Hamed Mohsenian-Rad. I hope to see some of you there. If you’re attending, please let me know so we can meet up! 🚀 #SGSMA2024 My dissertation focuses on enhancing situational awareness in power distribution systems through advanced sensor technologies like smart meters and distribution-level phasor measurement units (D-PMUs). I applied unsupervised machine learning to detect and classify unscheduled and infrequent events in the power grid. This involved analyzing time series data from smart meters and high-resolution D-PMUs (main focus of this research was on PMU data). I developed novel detection methods using generative adversarial networks and deep recurrent neural networks. Additionally, I proposed unsupervised clustering model based on graph representation learning that incorporates grid topology as well as the D-PMU time series, which we named it GraphPMU (You can find this journal paper in my google scholar). These methods enhance the accuracy of event detection, contributing to more resilient and efficient power distribution systems. For those who are interested, I attached my PhD dissertation ("Unsupervised Data-Driven Event Analysis of Smart Grid Time-Series"). P.S. This is the full name of the conference :D "International Conference on Smart Grid Synchronized Measurements and Analytics"

🎇 HAPPENING TOMORROW! Book now! 📅 EARLI SIG27 Satellite Series - Workshop: "Machine learning analysis for multimodal data", led by Dr. Indu Bodala (University of Southampton, UK). In this workshop, Indu will discuss her work on designing human-robot interactions for mental health and wellbeing applications. Human-robot interactions rely on inferring human intent from the data obtained during these interactions. Multimodal data refers to the data obtained from verbal and non-verbal communicative behaviours of humans recorded from the face, eyes, voice, body and even brain. She will discuss how analyzing data obtained from multiple modalities can enrich this inference and provide an interactive demonstration of machine learning techniques that can be used for this purpose. Additionally, there will be opportunities for hands-on practice at the techniques outlined.  BOOK NOW. Join us tomorrow! https://lnkd.in/en_qdU_i #machinelearning #multimodal #education #analysis

I am thrilled to announce that our latest research paper, "DDA-MLIC: Discriminator-Free Adversarial-Based Unsupervised Domain Adaptation for Multi-Label Image Classification," has been accepted at #WACV2024! 🎉 In this work, we tackle the challenge of #UnsupervisedDomainAdaptation (#UDA) for #MultiLabelImageClassification (#MLIC) by proposing a novel approach that eliminates the need for an additional discriminator subnet, which has been a major limitation in previous methods. Here’s a quick overview of our approach: ✨ DDA-MLIC introduces an adversarial critic directly derived from the task-specific classifier. ✨ We fit a two-component Gaussian Mixture Model (#GMM) on the source and target predictions to distinguish between two clusters, extracting a Gaussian distribution for each component. ✨ An adversarial loss based on the Fréchet distance is formulated using the resulting Gaussian distributions. ✨ Our method outperforms existing state-of-the-art techniques in terms of precision while using fewer parameters. We rigorously evaluated DDA-MLIC on several multi-label image datasets, demonstrating its effectiveness across three different types of domain shift. I invite you to check out 🔍: 📰 our paper: http://bit.ly/3USBna4 🔍 📊 our poster: https://bit.ly/3WQbXfH 🔍 👾 our publicly available code: https://lnkd.in/gm4mkp83. 🔍 We presented DDA-MLIC at #WACV2024. Feel free to reach out if you have any questions regarding our work. Enjie Ghorbel Anis Kacem Arunkumar Rathinam Djamila Aouada Our thanks to POST Luxembourg and Luxembourg National Research Fund (FNR) for their support! #ImageClassification #DomainAdaptation #ComputerVision #DeepLearning #MachineLearning #Research #WACV2024

Exciting news! Our paper on 'Finite Time Regret Bounds for Self-Tuning Regulation' has been accepted at the International Conference on Machine Learning (ICML) 2024 in Vienna, Austria. Our paper addresses the critical challenge of self-tuning regulation, which holds immense relevance across various industrial control systems. We introduce a reinforcement learning algorithm specifically tailored for this purpose and provide the first finite-time performance guarantees for tackling the self-tuning regulation problem. This advancement marks a significant stride towards the practical implementation of reinforcement learning algorithms in industrial control systems. Congratulations to my co-authors Dr. Rahul Singh, Avik Kar, and Prof. P. R. Kumar for their contributions! #ICML2024 #MachineLearning #ReinforcementLearning #SelfTuningRegulation #IndustrialControlSystems #ControlTheory #LearningTheory #TAMU #IISc

🚦 We are excited to share that our paper on "Traffic Light Control Using Reinforcement Learning" has been presented at the 2024 International Conference on Integrated Circuits and Communication Systems (ICICACS)! 📑 In urban areas, traffic congestion is a major issue affecting quality of life. Our study explores a novel approach using reinforcement learning to dynamically adjust traffic lights, reducing delays, and congestion, and improving safety and flow. 🔍 Dive into our research to learn how the Deep Q Learning algorithm optimizes traffic light sequences, cutting wait times by 50%. This innovative method holds promise for enhancing transport efficiency and sustainability. Check out the paper here: [https://lnkd.in/d2HPypDV] #TrafficControl #ReinforcementLearning #ICICACS2024 #TransportEfficiency #Sustainability 🚗