TCL Corporate Research(HK) Co., Ltd’s Post

🌟 Exciting News! 🌟 Thrilled to share that my paper, "WIP: AI-based Dynamic Joint Schedule Calculation for TSN over 5G using GCN-TD3," has been accepted at the IEEE International Conference on Emerging Technologies and Factory Automation #ETFA in Padova, Italy from 10th-13th September 2024. The research focuses on meeting hard real-time requirements for dynamic flows and topology changes in industrial settings. By co-configuring TSN-5G Time Aware Shaper (TAS) and 5G radio resource scheduling, we aim to enhance the dynamic scheduling of TSN-5G flows. The paper introduces a unified framework utilizing Graph Convolutional Networks (GCN) and the Twin Delayed Deep Deterministic Policy Gradient (TD3) Reinforcement Learning algorithm for dynamic flow scheduling. Looking forward to sharing insights and networking with fellow researchers at the conference! 🚀📊 #TSN #AI #5G #ETFA #conference

🤖💡 LLMs are great at reasoning! But how confident are they in their answers? Our latest survey dives into the world of LLM confidence estimation and calibration. 📰 https://lnkd.in/eHwKNe-4 Learn more about our #NAACL2024 paper in this post! What do we want 🎯? A perfectly calibrated model ⚖️! When it claims 70% confidence on 100 instances, exactly 70 predictions will be correct ✅ In practice, it is challenging! Why? 1️⃣ The generative space is vast!  2️⃣ Different token sequences may convey the same meaning. Mis-calibration can have bad consequences 🫣 Especially in critical areas like autonomous driving 🚗 or healthcare 💔 ⬆️ Overconfident models can make risky decisions  ⬇️ Underconfident ones can miss opportunities The confidence of sequence generation from LLMs can be measured from different sources of information. Teaching LLMs to express uncertainty (linguistic confidence) and combining different estimation methods is a growing area. Classification tasks also require calibration! Why? It has been widely observed that prompts and task formats such as the position bias in MCQA contain implicit biases 🧠💼 Our survey covers a large range of calibration methods ⚖️ Those methods are applied to both classification and generative tasks ☝️ 🤩 Future applications and exciting opportunities include:  💠 hallucination detection & mitigation 💠 ambiguity detection & selective generation 💠 uncertainty-guided data exploitation 💠 aligning the models' confidence levels with those of human users Here is the link to our survey paper: 📄 Paper: https://lnkd.in/eHwKNe-4  And consider following the authors 耿佳辉, Fengyu CAI, Yuxia Wang, Heinz Koeppl, Preslav Nakov and Iryna Gurevych of MBZUAI (Mohamed bin Zayed University of Artificial Intelligence), Ubiquitous Knowledge Processing (UKP) Lab and Computer Science, TU Darmstadt, if you are interested in more information or an exchange of ideas. See you this week in Mexico City 🇲🇽! #NLProc #NAACL2024 #Survey

🌟 Excited to announce my upcoming talk at ICC 2024 on June 11, 2024, from 8:30 PM - 10:00 PM, EEST (11:30 AM - 1:00 PM, MDT)! 🔍 Title: ORIENT: A Priority-Aware Energy-Efficient Approach for Latency-Sensitive Applications in 6G 🎙️ Abstract: As the era of 6G approaches, concerns about escalating energy consumption in computing and networking are on the rise. With an anticipated surge in connected devices and resource-intensive applications, ensuring sustainable resource allocation becomes paramount. Previous discussions on resource allocation strategies have often focused on single-domain orchestration, overlooking the unique challenges posed by 6G. To bridge this gap, we delve into the joint problem of service instance placement and assignment, path selection, and request prioritization, termed PIRA. Our objective? Maximizing overall system profit while minimizing energy consumption over an extended period, all while considering end-to-end latency requirements and resource capacity constraints. Leveraging queuing theory to estimate the Age of Information (AoI) for requests, we tackle this complex issue head-on, proving its NP-hardness and introducing ORIENT—a method grounded in Double Dueling Deep Q-Learning (D3QL) and Graph Neural Networks (GNNs) for state encoding. Our extensive simulations showcase ORIENT's ability to generate near-optimal solutions across varying system sizes and request counts. 👉 Join me to explore how ORIENT pioneers an energy-efficient path in the 6G landscape, optimizing resource allocation while meeting the demands of latency-sensitive applications. For more details, visit here: https://lnkd.in/dYzq9hED #6G #AI #ML #GNN #Paper #Conference #Energy_Efficiency

#cvpr2024 #wicv2024 🌟 Excited to announce that our paper "GestFormer: Multiscale Wavelet Pooling Transformer Network for Dynamic Hand Gesture Recognition" will be presented at #CVPR24! 🌟 workshop for women in computer vision 🕒 Join us at the poster session from 12:30 PM to learn about our innovative approach to gesture recognition using a resource-efficient transformer model. Key highlights: - Proposes GestFormer, a novel architecture for dynamic hand gesture recognition. - Utilizes PoolFormer for efficient token mixing and wavelet transform for space-invariant feature extraction. - Achieves superior performance with fewer parameters on datasets like NVidia Dynamic Hand Gesture and Briareo. - Experiments were conducted on multiple modalities, including infrared, depth, and optical flow. We invite you to explore the details of our work and engage with us during the session. Read the full paper here: [GestFormer: Multiscale Wavelet Pooling Transformer Network for Dynamic Hand Gesture Recognition](https://lnkd.in/gqWYdWtP) #CVPR2024 #AI #MachineLearning #DeepLearning #GestureRecognition #TransformerModels #Research #Innovation Mallika Garg, Debashis Ghosh, Pyari Mohan Pradhan

🌟 Congratulations to NTU MAE’s Prof. Yang Jianfei and his research team for Winning Top Honors at ECCV 2024! 🌟 🎉 Thrilled to share our team's achievement, led by Prof. Jianfei Yang, of winning 1st place and the Best Paper Award at the Data Distillation Challenge at the top-tier computer vision conference, European Conference on Computer Vision (ECCV-2024, Milan)! 🎉 Their award-winning paper, "Enhancing Dataset Distillation via Label Inconsistency Elimination and Learning Pattern Refinement", explores Dataset Distillation (DD)—a transformative approach that compresses large datasets into compact, synthetic versions, making AI training on edge devices more feasible. This advancement is crucial for applications with limited computational power, such as robotics and industrial automation. Their method, Modified Difficulty-Aligned Trajectory Matching (M-DATM), enhances DD by resolving label inconsistencies and refining learning patterns. The result? High model performance on smaller datasets with minimal performance loss, paving the way for impactful AI deployment in constrained environments. A huge congratulations to our team for this outstanding work pushing the boundaries of AI! 🚀👏 Chuhao Zhou Haozhi Cao #NTUsg #NTUMAE #ECCV2024 #DataDistillation #AIResearch #Robotics #Innovation #AI #IndustrialAutomation #Automation #Achievement

🚀 ISPC 2024: Day 2 Presentation of Break-Out Sessions 🚀 Today at ISPC 2024, we dove deep into the future of #DigitalManagement and AI! 🌐✨ Dr. Anton Ravindran, President of the Alliance for AI & Humanity, joined us virtually from Singapore to share insights on the exponential growth of AI. He discussed how Generative AI is revolutionizing industries by outperforming humans in coding, design, and medical analysis, driven by an explosion in data and computing power. 📈🤖 However, with great power comes great responsibility! 🌱💡 Dr. Ravindran also warned about the ethical challenges AI poses, including bias, misinformation, and the need for responsible development. He emphasized the urgent need for regulatory frameworks to guide this fast-growing tech. 🛡️⚖️ Stay tuned for Part 2, where we dive into the exciting breakout sessions and more! 🧠💬 #ISPC2024 #ArtificialIntelligence #DigitalManagement #AIForGood #TechEthics #FutureOfAI #Innovation #TechConference Richard C. Geibel

An important part of the international #ISPC2024 conference were the valuable and enriching keynote speeches and welcome addresses. We were particularly pleased to receive such a well-founded presentation 👌 from Dr. Anton Ravindran, who, in his remarks from Singapore, addressed the great importance of a well-balanced and finely tuned interplay between #AI and #humanity. Once again, we would like to express our sincere thanks 👍 for these important remarks. #digitalization #ethics E-Commerce Institut

C3 Technology Advisors AIaaS KickaaS Tip of the week: A.I. for the win! Want to get ahead of your competition? Sly Rye the C3 UC & AI guy explains how to get in on the best A.I. conference in the region. And it is free to attend. Here's the scoop.

 Neuralink introduces the first commercial trial of Brain-Computer Interface (BCI) application on humans by a private company: https://lnkd.in/e8UkWnct This marks a significant milestone, although BCI technology itself is not entirely new. A decade ago, I encountered the book "I, Cyborg" authored by former cybernetics professor Warwick, who utilized BCI on himself. The book opens with the profound statement, "I was born human." In 1998, he even had an RFID transmitter implanted beneath his skin: https://lnkd.in/eGfGfH35 Experiments documenting communication between the nervous system and electronic devices in animals date back to the 1970s: https://lnkd.in/euA3bhTB Elon Musk is now advancing what is essentially the current state-of-the-art in science and technology. While it may not represent the most groundbreaking technological leap, its potential societal impact is unparalleled. BCI holds the promise not only to restore lost capabilities due to accidents but also to expand our mental, sensory, and physical capacities.

Retrieval augmented generation is a technique for enhancing the accuracy and reliability of #generativeAI models. Start building applications that deliver serious #LLM-based RAG applications at #GTC24 🗓️ Add these #RAG training sessions to your calendar: https://nvda.ws/48UqtpQ