Tom Yeh

New Publication This study explores advanced photoactive CO-releasing materials using heterocyclic structures. It reveals strong Mn(CO)₃-heterocycle interactions in ground states and weakened CO binding in excited states, facilitating CO release. Quantum chemical and AIM analyses confirm changes in electron density and bond elongation, highlighting the potential of these molecules for efficient CO delivery. The work was in collaboration with Anik Sen GITAM Deemed University #researchinGITAM #GITAM #CTC #ComputationlChemistry

11

2 Comments

How can AI rapidly generate diverse solutions to meet a design target—without relying on "deep" generative models? Our latest publication, "Generative Inverse Design of Metamaterials with Functional Responses by Interpretable Learning," tackles this challenge by proposing a new method for the rapid inverse design of metamaterials that meet target functional behaviors, under the constraint of small data. In this work, we introduce 𝗥𝗜𝗚𝗜𝗗 (𝗥𝗮𝗻𝗱𝗼𝗺-𝗳𝗼𝗿𝗲𝘀𝘁-𝗯𝗮𝘀𝗲𝗱 𝗜𝗻𝘁𝗲𝗿𝗽𝗿𝗲𝘁𝗮𝗯𝗹𝗲 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗶𝘃𝗲 𝗜𝗻𝘃𝗲𝗿𝘀𝗲 𝗗𝗲𝘀𝗶𝗴𝗻), a single-shot inverse design approach for generating diverse metamaterials that meet the target response, like bandgaps. Unlike commonly used, data-intensive deep-learning models, RIGID emphasizes small data efficiency, interpretability, and broader design space exploration. Key Highlights: • 𝗦𝗮𝗺𝗽𝗹𝗲 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆: Achieves desired designs with fewer than 250 training samples, ideal for scenarios with limited data. • 𝗜𝗻𝘁𝗲𝗿𝗽𝗿𝗲𝘁𝗮𝗯𝗶𝗹𝗶𝘁𝘆: Estimates a likelihood of target satisfaction for any design, enabling informed decision-making and facilitating exploration-exploitation trade-off. • 𝗦𝗼𝗹𝘂𝘁𝗶𝗼𝗻 𝗗𝗶𝘃𝗲𝗿𝘀𝗶𝘁𝘆: Discovers diverse solutions for the same target, enabling consideration of additional performance metrics. We validated RIGID on acoustic metamaterial and optical metasurface designs, demonstrating its power to enable on-demand generative inverse design under limited computational resources. Check out the full article: https://lnkd.in/gtVnmw8Q Try out the code: https://lnkd.in/gHy76wfQ #Metamaterials #InverseDesign #GenerativeDesign #MachineLearning #Interpretability

109

4 Comments

📢 Excited to share recent research achievements with my PhD student Reza Vafaee Here are three of our latest papers: 1- Learning-based Sparse Sensing with Performance Guarantees in IEEE Transactions on Automatic Control A novel method for converting weighted sensor schedules to sparse ones while maintaining performance. Read more: https://lnkd.in/e4eD2QGF 2- Real-time Sensor Selection for Time-Varying Networks with Guaranteed Performance in Automatica Dynamic sensor selection for LTV systems with high performance and minimal sensors. Read more: https://lnkd.in/eCYTmb9c 3- Exploring Non-Submodular Scheduling for Large-Scale Sensor Networks in IEEE Control Systems Letters An effective greedy algorithm for non-submodular sensor scheduling. Read more: https://lnkd.in/etxfgJaS Proud of our progress in sparse sensing in complex networks! 🚀 #Research #SensorNetworks #ControlSystems #SparseSensing #IEEE #Automatica #PhDResearch

64

4 Comments

Our paper titled "AdaFedProx: A Heterogeneity-aware Federated Deep Reinforcement Learning for Medical Image Classification" got accepted in IEEE Transactions on Consumer Electronics (impact factor : 4.3). Authors: Pranab Sahoo, Ashutosh Tripathi, Sriparna Saha, Samrat Mondal, Jyoti Prakash S., Bhisham Sharma Abstract: This research introduces a novel heterogeneity-aware federated learning approach named AdaFedProx to address the performance degradation resulting from heterogeneity. AdaFedProx utilizes a reinforcement learning-driven strategy, introducing a tailored proximal term into the objective function to incorporate partial client information in heterogeneous settings. Dynamically determining the proximal term based on individual client states, including critical characteristics like data distribution, system specification, and performance feedback, empowers AdaFedProx to make regularized decisions throughout the federated training. AdaFedProx consistently outperforms four state-of-the-art methods on three medical datasets across diverse heterogeneity settings (0%, 50%, and 75% straggler). Heartiest congratulations to all authors!

276

15 Comments

📣 📣 📣 We have extended the deadline for AAAI workshop on Planning in the Era of LLMs (LM4Plan @ AAAI 2025)! 📣 📣 📣 https://lnkd.in/eP5yQHkf The new deadline is *November 28* AOE! Large Language Models (LLMs) are a disruptive force, changing how research was done in many sub-areas of AI. Planning is one of the last bastions that remain standing. The focus of this workshop is on the questions in the intersection of these areas. Some of the specific areas we would like to gain a better understanding in include: what LLMs can contribute to planning, how LLMs can/should be used, what are the pitfalls of using LLMs, what are the guarantees that can be obtained. #planning #LLM #agents Sarath Sreedharan Jiayuan Mao Subbarao Kambhampati @Wenlong Huang

33

#RENCI researchers Anirban Mandal + Cong W. are listed as authors on a paper being presented at 2 PM ET in Room B311 at #SC24. https://lnkd.in/guG7Np4E #LLM #anomalydetection #computationalworkflows #computationalefficiency

11

Our paper, "On the Effect of Quantization on Dynamic Mode Decomposition," has been accepted for presentation at IEEE CDC 2024. Dynamic Mode Decomposition (DMD) is a widely used data-driven algorithm for estimating the Koopman Operator. The paper delves into how the estimation process is influenced by data quantization, uncovering a key theoretical finding that enables the accurate recovery of unquantized estimates from quantized data. This analysis marks the first exploration of the impact of quantization on data-driven algorithms, offering new insights into how communication constraints can affect learning algorithms, particularly in the realm of machine learning. Explore the full preprint at https://lnkd.in/gMrj6q5i. Gratitude to Dipankar Maity for conceptualizing the problem and Sriram Narayanan for the codebase.

90

2 Comments

My Machine Learning course in Bengali language will be running once again on NPTEL, during June-September 2024! Please join in large numbers, and also spread the word in your circles! https://lnkd.in/drm8UtuJ

137

2 Comments

JEE Advanced 2018 Physics MCQ With one or more correct options MCQ on Circular Motion of particle https://lnkd.in/gjD4bsyq #JEEAdvancedPhysicsMCQ #CircularMotionOfGivenParticle #DeterminationOfSpeed #AndAngularMomentumOfParticle

11

11 Comments

INTEGRAL CALCULUS; DEFINITE INTEGRALS; VOLUME OF INFINITE SOLID; REDUCIBILITY OF A POLYNOMIAL JEE ADVANCE #JEEMADEEASYKOTA, #MATHSMADEEASYKOTA, #NDAMADEEASYKOTA, #CETMADEEASYKOTA

1

Perturbation theory for your arbitrary wavefunctions: https://lnkd.in/egaKPQw3 Brought to you by LOKHANDE RUGWED and Pratiksha Gaikwad #computationalchemistry #perturbationtheory #coupledcluster

39

Multidirectional Associative Memory. In our collaborative work with Sahibia Kaur Vohra, PhD student and Dr Devarshi Das, Assistant Professor at IIT Ropar, we demonstrate the implementation and application of Bio-inspired spiking Multidirectional Associative Memory. We are happy to share our latest publication in IEEE Transactions on Circuits and Systems I: Regular Papers, titled "SpiMAM: CMOS Implementation of Bio-Inspired Spiking Multidirectional Associative Memory Featuring In-Situ Learning." 🔍 Abstract: Associative Memory (AM) is a powerful mechanism for retrieving information from partial data. In our study, we introduce SpiMAM, a spiking neural network (SNN)-based Multidirectional Associative Memory Neural Network (MAMNN), which offers greater bio-plausibility, sparsity, and message storage capacity compared to traditional artificial neural networks (ANNs). Unlike previous implementations, SpiMAM employs in-situ synaptic training using CMOS circuits with spike-timing-dependent plasticity (STDP). 🔧 Key Features: --Bio-inspired spiking MAMNN with a winner-take-all mechanism --CMOS circuit design of memristive synapse featuring STDP --On-chip synaptic weight training without additional computing platforms --Transistor-level implementation in 180 nm standard CMOS technology --Robustness demonstrated through post-layout simulations against PVT, mismatch variation, pixel flip, hard faults, memristive drifts, and Gaussian noise --Significant reduction in synapses (86%) and neurons (70%) for pattern recognition tasks This work paves the way for more efficient and robust pattern recognition applications. 🔗 Read the full paper: https://lnkd.in/gfPR3aCT Authors: S. K. Vohra, M. Sakare, A. P. James, and D. M. Das #Research #Innovation #NeuralNetworks #SpikingNeuralNetworks #AssociativeMemory #CMOS #STDP #IEEE #PatternRecognition #BioInspiredComputing

64

4 Comments

Glad to share that our paper “Task Scheduling in Cognitive Multifunction Radar using Model-based DRL” is published in IEEE Transactions on Aerospace and Electronic Systems. The work is supervised by Prof Ravi Adve and is a collaborative work with Dr Jack Ding and Dr Peter W Moo of Defence Research and Development Canada Ottawa Research Centre (DRDC Ottawa), the funding body. Here, we address the need for online task scheduling in a cognitive multifunction radar (MFR) without prior knowledge of the environment. We draw inspiration from recent advancements in model-based deep reinforcement learning (DRL), specifically the applicability of MuZero, to enable cognitive MFR in unknown and continuously changing environments. In this approach, the scheduler learns an abstract Markov decision process (MDP) model of the environment, allowing near-optimal abstract MDP planning to translate effectively into the actual environment. In this paper, we modify the original MuZero algorithm to accommodate large number of tasks by incorporating prior knowledge of the task scheduling problem. Our numerical results show that the modified MuZero approach is effective and computationally efficient in complex radar scenarios. The paper is available here https://lnkd.in/gyryx5ga #cognitiveradar #artificialintelligence #drdc #uoft #ieee

115

34 Comments

Beautiful: The arrival of bacterial AI (bANN) in full swing. Professor Sangram Bagh of Saha Institute of Nuclear Physics (SINP),Kolkata and his team has demonstrated the following (in his own words): "Genetically engineered bacteria can identify prime numbers, vowels, and even determine the maximum number of pieces of pizza or pie that can be obtained from a given number of straight cuts. In addition, the genetically modified bacteria are able to answer mathematical questions such as whether a number n's factorial is divisible by n × (n + 1)/2 OR whether a number n's square can be expressed as the sum of three factorials." Office of the Principal Scientific Adviser to the Government of India CM Office, GoUP DNA Data Storage Alliance Synthetic Biology DNA Storage Lab @ Technion #DNAcomputing #Bacterialcomputing #Bacterial AI, #Naturalcomputing #INDIA #MakeinIndia Read the full paper at https://lnkd.in/d8ffaQFf

57

Upcoming #tutorial by #impactlabasu's Ayan Banerjee and Sandeep K. S. Gupta with Marzia Cescon at IEEE EMBS BHI 2024 https://bhi.embs.org/2024/ in Houston TX Tools and techniques for learning #digitaltwins for #biomedical applications Sunday | November 10, 2024 | 9:00 – 12:30 Digital twins are an exciting new area with applications in many biomedical fields including #precisionmedicine, generation of patient specific data to aid in AI for health, and large scale in-silico testing of devices and drugs. However, a serious bottleneck in real world usage of digital twins is their fidelity. Can digital twins be accurately learned from real-world data? This question is at the center of any attempt at using digital twins in medicine and health. In this tutorial, we take a deep dive into the challenges of learning digital twins from real world data, which include digital twin structures, first principle based #modelrecovery, data scarcity, data heterogeneity, data privacy and data requirement trade-off, learning from incomplete knowledge, and generalization across domains. We then discuss tools and techniques both state-of-art and cutting-edge novel research (topic of recently awarded #NSF grant), to address these challenges. This tutorial directly addresses the conference theme of #DeepMedicine and #AIforHealth.

13

Every once in a while I see a piece of data that clearly illustrates why a trend may be happening. Today I was investigating what are the Flagship conferences for the IEEE Circuits and Systems Society to determine if my North American (IEEE Regions 1-7 and part of 9) students are eligible to apply for a travel grant. Of course, many have noted that the field of circuits in academic research is in serious jeopardy in North America (IEEE Regions 1-7 and part of 9). Of course in reviewing where the flagship conferences are this year one will note that none of the #CASS flagship conferences occur in North America. Essentially, an IEEE society doesn't find it important to hold even a single flagship regional conference in North America! As of the 2022 report, North America has 2x as many members as the next largest region. https://lnkd.in/d8GMbtQA

27

3 Comments

🚀 Exciting News Alert! 🚀 👏 A paper titled as "CMOT: A Cross-Modality Transformer for RGB-D Fusion in Person Re-Identification with Online Learning Capabilities" which I co authored (Supervised also) has been published in the prestigious journal 'Knowledge Based Systems', ranking in the top 10% of journals in the realm of big data and artificial intelligence with an impact factor of 8.8 (Q1, Top 10%). Let's give a round of applause to Hamza Mukhtar, a rising star in my research lab! 👏 Hamza, is a dedicated MS student at Department of Computer Science, UET Lahore and a brilliant Research Officer at the Intelligent Criminology Research Lab NCAI - UET Lahore under my mentorship, has just achieved a remarkable milestone. This study makes significant strides in the field of multi-modality reID, contributing to advancements in several key areas: 1️⃣ Introduces CMOT, a transformer-based multi-modality reID model, revolutionizing the concurrent extraction of spatiotemporal information and cross-modality fusion. This framework utilizes both RGB and depth modalities, harnessing dual-stream CNNs and a fusion transformer to leverage complementary and discriminative features from dynamic sequence inputs. 2️⃣ Introduces SAMR and CAMI, driven by self-attention and cross-attention mechanisms, respectively. These components enhance semantic information within each modality and facilitate cross-modal feature interaction, resulting in efficient feature-level multimodal fusion. The incorporation of BEFFB further improves the model's representation capability while reducing computational complexity. 3️⃣ Pioneers a novel approach by proposing a similarity distance measure-based triplet loss function, transforming CMOT into a few-shot network with online learning capabilities for reID inference. This enables incremental updates and adaptation to new, unseen data without the need for extensive model retraining. https://lnkd.in/dYb2Fc2w #Artificial_Intelligence, #ComputerVision, #DeepLearning, #AI, #ReIdentification, #Transformers Seventh Reality Fukatsoft Services

101

8 Comments

Excited to share our recently published paper in IEEE Transactions on Control of Network Systems: "Distributed Optimization for Quadratic Cost Functions with Quantized Communication and Finite-Time Convergence". We propose two novel distributed algorithms for solving optimization problems over large-scale networks: a synchronous and an asynchronous. Both operate with quantized values, achieve finite-time convergence, and include distributed termination mechanisms.  We demonstrate applications in optimal task scheduling for data centers and global model aggregation for distributed federated learning. Special thanks to my collaborators Andreas Grammenos, Eva Kalyvianaki, Christoforos Hadjicostis, Themistoklis Charalambous, and Karl Henrik Johansson on this research. Read the full paper:  https://lnkd.in/dg4P9JWQ  and  https://lnkd.in/dip69Am8 #DistributedOptimization #NetworkSystems #DataCenters #FederatedLearning #IEEE

45

6 Comments