Showing papers by "Saibal Mukhopadhyay published in 2022"
TL;DR: In this paper , an Artificial Intelligence (AI) model was used to identify time domain heart rate variability (HRV) measures most suitable for short term ECG in COVID-19 recovered subjects.
Abstract: Cardiovascular dysautonomia comprising postural orthostatic tachycardia syndrome (POTS) and orthostatic hypotension (OH) is one of the presentations in COVID-19 recovered subjects. We aim to determine the prevalence of cardiovascular dysautonomia in post COVID-19 patients and to evaluate an Artificial Intelligence (AI) model to identify time domain heart rate variability (HRV) measures most suitable for short term ECG in these subjects.This observational study enrolled 92 recently COVID-19 recovered subjects who underwent measurement of heart rate and blood pressure response to standing up from supine position and a 12-lead ECG recording for 60 s period during supine paced breathing. Using feature extraction, ECG features including those of HRV (RMSSD and SDNN) were obtained. An AI model was constructed with ShAP AI interpretability to determine time domain HRV features representing post COVID-19 recovered state. In addition, 120 healthy volunteers were enrolled as controls.Cardiovascular dysautonomia was present in 15.21% (OH:13.04%; POTS:2.17%). Patients with OH had significantly lower HRV and higher inflammatory markers. HRV (RMSSD) was significantly lower in post COVID-19 patients compared to healthy controls (13.9 ± 11.8 ms vs 19.9 ± 19.5 ms; P = 0.01) with inverse correlation between HRV and inflammatory markers. Multiple perceptron was best performing AI model with HRV(RMSSD) being the top time domain HRV feature distinguishing between COVID-19 recovered patients and healthy controls.Present study showed that cardiovascular dysautonomia is common in COVID-19 recovered subjects with a significantly lower HRV compared to healthy controls. The AI model was able to distinguish between COVID-19 recovered patients and healthy controls.
26 citations
Proceedings Article•
13 citations
TL;DR: In this paper , the authors proposed individualized aggressive LDL-C goals after acute coronary syndrome (ACS), which can be rapidly achieved with high intensity statin therapy and subsequent goal-directed adjunctive treatment with ezetimibe and PCSK9 inhibitors.
5 citations
13 Jul 2022
TL;DR: A novel Hebbian learning method to extract the global feature of a point set in StarCraft II game units, and its application to predict the movement of the points and introduces the concept of neuron activity aware learning combined with k-Winner-Takes-All.
Abstract: Learning the evolution of real-time strategy (RTS) game is a challenging problem in artificial intelligent (AI) system. In this paper, we present a novel Hebbian learning method to extract the global feature of a point set in StarCraft II game units, and its application to predict the movement of the points. Our model includes encoder, LSTM, and decoder, and we train the encoder with the unsupervised learning method. We introduce the concept of neuron activity aware learning combined with k-Winner-Takes-All. The optimal value of neuron activity is mathematically derived, and experiments support the effectiveness of the concept over the downstream task. Our Hebbian learning rule benefits the prediction with lower loss compared to self-supervised learning. Also, our model significantly saves the computational cost such as activations and FLOPs compared to a frame-based approach.
4 citations
TL;DR: In this article , the authors have developed a consensus statement to provide guidance for management of diabetic dyslipidemia in very high risk population, which is based on a series of 165 webinars conducted by the Lipid Association of India across the country from May 2020 to July 2021, involving 155 experts in endocrinology and cardiology and an additional 2880 physicians.
3 citations
TL;DR: The proposed architecture is used to accelerate a hybrid SNN architecture that couples off-chip supervised and on-chip unsupervised (STDP) training and the accuracy of this hybrid network is 10.84% higher than STDP trained accuracy result and 1.4% higher compared to the backpropagated training-based ConvSNN result with the CIFAR-10 dataset.
Abstract: We present a processing-in-memory (PIM)-based hardware platform, referred to as MONETA, for on-chip acceleration of inference and learning in hybrid convolutional spiking neural network. MONETAuses 8T static random-access memory (SRAM)-based PIM cores for vector matrix multiplication (VMM) augmented with spike-time-dependent-plasticity (STDP) based weight update. The spiking neural network (SNN)-focused data flow is presented to minimize data movement in MONETAwhile ensuring learning accuracy. MONETAsupports on-line and on-chip training on PIM architecture. The STDP-trained convolutional neural network within SNN (ConvSNN) with the proposed data flow, 4-bit input precision, and 8-bit weight precision shows only 1.63% lower accuracy in CIFAR-10 compared to the STDP accuracy implemented by the software. Further, the proposed architecture is used to accelerate a hybrid SNN architecture that couples off-chip supervised (back propagation through time) and on-chip unsupervised (STDP) training. We also evaluate the hybrid network architecture with the proposed data flow. The accuracy of this hybrid network is 10.84% higher than STDP trained accuracy result and 1.4% higher compared to the backpropagated training-based ConvSNN result with the CIFAR-10 dataset. Physical design of MONETAin 65 nm complementary metal-oxide-semiconductor (CMOS) shows 18.69 tera operation per second (TOPS)/W, 7.25 TOPS/W and 10.41 TOPS/W power efficiencies for the inference mode, learning mode, and hybrid learning mode, respectively.
3 citations
TL;DR: In this article , a heterogeneous recurrent spiking neural network (HRSNN) with unsupervised learning for spatio-temporal classification of video activity recognition tasks was proposed.
Abstract: Spiking Neural Networks are often touted as brain-inspired learning models for the third wave of Artificial Intelligence. Although recent SNNs trained with supervised backpropagation show classification accuracy comparable to deep networks, the performance of unsupervised learning-based SNNs remains much lower. This paper presents a heterogeneous recurrent spiking neural network (HRSNN) with unsupervised learning for spatio-temporal classification of video activity recognition tasks on RGB (KTH, UCF11, UCF101) and event-based datasets (DVS128 Gesture). We observed an accuracy of 94.32% for the KTH dataset, 79.58% and 77.53% for the UCF11 and UCF101 datasets, respectively, and an accuracy of 96.54% on the event-based DVS Gesture dataset using the novel unsupervised HRSNN model. The key novelty of the HRSNN is that the recurrent layer in HRSNN consists of heterogeneous neurons with varying firing/relaxation dynamics, and they are trained via heterogeneous spike-time-dependent-plasticity (STDP) with varying learning dynamics for each synapse. We show that this novel combination of heterogeneity in architecture and learning method outperforms current homogeneous spiking neural networks. We further show that HRSNN can achieve similar performance to state-of-the-art backpropagation trained supervised SNN, but with less computation (fewer neurons and sparse connection) and less training data.
3 citations
TL;DR: In this article , the role of novel lipid biomarkers and their genetic polymorphisms in young (<50 years) ST-segment elevation myocardial infarction (STEMI) patients was investigated.
Abstract: There is an increasing prevalence of coronary artery disease (CAD) in younger individuals. Novel lipid biomarkers such as lipoprotein-a (Lp-a), Apo A1, Apo B and paraoxanase-1 (PON1) serve as important risk predictors for development of CAD. There is little evidence regarding the role of novel lipid biomarkers and their genetic polymorphisms in young (<50 years) ST-segment elevation myocardial infarction (STEMI) patients.
2 citations
TL;DR: Bmpedoic acid is a welcome addition to the existing non-statin therapies such as ezetimibe, bile acid sequestrants, and PCSK9 inhibitors and has a low frequency of muscle-related side effects, minimal drug interactions, and a significant reduction in high-sensitivity C-reactive protein (hsCRP) make it a useful adjunct for LDL-C lowering.
Abstract: Lipid-lowering therapy plays a crucial role in reducing adverse cardiovascular (CV) events in patients with established atherosclerotic cardiovascular disease (ASCVD) and familial hypercholesterolemia. Lifestyle interventions along with high-intensity statin therapy are the first-line management strategy followed by ezetimibe. Only about 20-30% of patients who are on maximally tolerated statins reach recommended low-density lipoprotein cholesterol (LDL-C) goals. Several factors contribute to the problem, including adherence issues, prescription of less than high-intensity statin therapy, and de-escalation of statin dosages, but in patients with very high baseline LDL-C levels, including those with familial hypercholesterolemia and those who are intolerant to statins, it is critical to expand our arsenal of LDL-C-lowering medications. Moreover, in the extreme risk group of patients with an LDL-C goal of ≤30 mg/dL according to the Lipid Association of India (LAI) risk stratification algorithm, there is a significant residual risk requiring the addition of non-statin drugs to achieve LAI recommended targets. This makes bempedoic acid a welcome addition to the existing non-statin therapies such as ezetimibe, bile acid sequestrants, and PCSK9 inhibitors. A low frequency of muscle-related side effects, minimal drug interactions, a significant reduction in high-sensitivity C-reactive protein (hsCRP), and a lower incidence of new-onset or worsening diabetes make it a useful adjunct for LDL-C lowering. However, the CV outcomes trial results are still pending. In this LAI consensus document, we discuss the pharmacology, indications, contraindications, advantages, and evidence-based recommendations for the use of bempedoic acid in clinical practice.
2 citations
18 Jul 2022
TL;DR: ModelNet is an Artificial Neural Network that can estimate spatial/semantic model uncertainties of a DNN based object detection with less computation overhead and a case study of uncertainty driven adaptive sensor using ModelNet is presented.
Abstract: Quantifying model uncertainty of Deep Neural Network (DNN) is important to understand the reliability of the model prediction and avoid risks in safety critical applications. Various approaches, including Bayesian neural networks, Monte-Carlo dropout, and ensembles, are suggested to measure the model uncertainty; but with huge computational cost. We present ModelNet, an Artificial Neural Network (ANN) that can estimate spatial/semantic model uncertainties of a DNN based object detection with less computation overhead. ModelNet is a deterministic ANN that distills the predictive distribution of stochastic DNN. Experimental results show that ModelNet can learn the uncertainty estimation from stochastic DNN in various architectures. ModelNet can perform as a probabilistic object detector with 39x-179x less number of operations, or as an uncertainty assistant to a task network with 1.4x more parameters and 38x less number of operations compared to stochastic DNN. Moreover, a case study of uncertainty driven adaptive sensor using ModelNet is presented.
1 citations
TL;DR: This paper proposes a deep learning model that learns to predict unknown spatio-temporal dynamics using data from sparsely-distributed data sites using the radial basis function (RBF) collocation method which is often used for meshfree solution of partial differential equations.
Abstract: In this paper, we address the problem of predicting complex, nonlinear spatio-temporal dynamics when available data are recorded at irregularly spaced sparse spatial locations. Most of the existing deep learning models for modelling spatio-temporal dynamics are either designed for data in a regular grid or struggle to uncover the spatial relations from sparse and irregularly spaced data sites. We propose a deep learning model that learns to predict unknown spatio-temporal dynamics using data from sparsely-distributed data sites. We base our approach on the radial basis function (RBF) collocation method which is often used for meshfree solution of partial differential equations. The RBF framework allows us to unravel the observed spatio-temporal function and learn the spatial interactions among data sites on the RBF-space. The learned spatial features are then used to compose multilevel transformations of the raw observations and predict its evolution in future time steps. We demonstrate the advantage of our approach using both synthetic and real-world climate data. This article is part of the theme issue ‘Data-driven prediction in dynamical systems’.
TL;DR: In this paper , the authors investigated the performance of three different OAT methods: (1) model-based methods (i.e., classical ray-based OAT using a linearized forward model), (2) data-driven methods (such as deep learning) to directly learn the inverse model, and (3) a hybrid solution, which combines deep learning of the forward model with a standard recursive optimization to estimate SSPs.
Abstract: Underwater sound propagation is primarily driven by a nonlinear forward model relating variability of the ocean sound speed profile (SSP) to the acoustic observations (e.g., eigenray arrival times). Ocean acoustic tomography (OAT) methods aim at reconstructing SSP variations (with respect to a reference environment) from changes of the acoustic measurements between multiple source-receiver pairs. This article investigates the performance of three different OAT methods: (1) model-based methods (i.e., classical ray-based OAT using a linearized forward model), (2) data-driven methods (such as deep learning) to directly learn the inverse model, and (3) a hybrid solution [i.e., the neural adjoint (NA) method], which combines deep learning of the forward model with a standard recursive optimization to estimate SSPs. Additionally, synthetic SSPs were generated to augment the variability of the training set. These methods were tested with modeled ray arrivals calculated for a downward refracting environment with mild fluctuations of the thermocline. Idealized towed and fixed source configurations are considered. Results indicate that merging data-driven and model-based methods can benefit OAT predictions depending on the selected sensing configurations and actual ray coverage of the water column. But ultimately, the robustness of OAT predictions depends on the dynamics of the SSP variations.
DOI•
19 Jun 2022
TL;DR: Low-latency and high-throughput, configurable architecture for computing sparse Finite Impulse Response in real-time Radio Frequency domain is proposed, which supports configurability in filter tap locations and handling of locally dense taps, making it more adaptable to Radio Frequency environments.
Abstract: $A$ low-latency and high-throughput, configurable architecture for computing sparse Finite Impulse Response in real-time Radio Frequency domain is proposed. The massively parallel architecture uses distributed control in association with near-memory techniques to optimize area and power. It supports configurability in filter tap locations and handling of locally dense taps, making it more adaptable to Radio Frequency environments.
TL;DR: This work derives a data-movement-aware weight selection method that does not require retraining to preserve its original performance and can reduce the accuracy degradation from 60 - 90% (without protection) to 1 - 2% for different DNNs across diverse datasets.
Abstract: In recent years, processing in memory (PIM) based mixed-signal designs have been proposed as energy- and area-efficient solutions with ultra high throughput to accelerate DNN computations. However, PIM designs are sensitive to imperfections such as noise, weight and conductance variations that substantially degrade the DNN accuracy. To address this issue, we propose a novel algorithm-hardware co-design framework hereafter referred to as HybridAC that simulta-neously avoids accuracy degradation due to imperfections, improves area utilization, and reduces data movement and energy dissipation. We derive a data-movement-aware weight selection method that does not require retraining to preserve its original performance. It computes a fraction of the results with a small number of variation-sensitive weights using a robust digital accelerator, while the main computation is performed in analog PIM units. This is the first work that not only provides a variation-robust architecture, but also improves the area, power, and energy of the existing designs considerably. HybridAC is adapted to leverage the preceding weight selection method by reducing ADC precision, peripheral circuitry, and hybrid quantization to optimize the design. Our comprehensive experiments show that, even in the pres-ence of variation as high as 50%, HybridAC can reduce the accuracy degradation from 60 - 90% (without protection) to 1 - 2% for different DNNs across diverse datasets. In addition to providing more robust computation, compared to the ISAAC (SRE), HybridAC improves the execution time, energy, area, power, area-efficiency, and power-efficiency by 26%(14%), 52%(40%), 28%(28%), 57%(45%), 43%(5 × ), and 81%(3.9 × ), respectively.
18 Jul 2022
TL;DR: A novel Recurrent Graph Network approach for predicting discrete marked event sequences by learning the underlying complex stochastic process by changing the self-attention mechanism from attending over past events to attending over event types is presented.
Abstract: We present a novel Recurrent Graph Network (RGN) approach for predicting discrete marked event sequences by learning the underlying complex stochastic process. Using the framework of Point Processes, we interpret a marked discrete event sequence as the superposition of different sequences each of a unique type. The nodes of the Graph Network use LSTM to incorporate past information whereas a Graph Attention Network (GAT Network) introduces strong inductive biases to capture the interaction between these different types of events. By changing the self-attention mechanism from attending over past events to attending over event types, we obtain a reduction in time and space complexity from $\mathcal{O}(N^{2})$ (total number of events) to $\mathcal{O}(\vert \mathcal{Y}\vert^{2})$ (number of event types). Experiments show that the proposed approach improves performance in log-likelihood, prediction and goodness-of-fit tasks with lower time and space complexity compared to state-of-the art Transformer based architectures.
TL;DR: Dysglycemia was associated with worse clinical outcomes at 30 days, and use of established pharmacotherapeutic risk-reduction strategies among patients with known diabetes was rare, highlighting missed opportunities for screening and management of dys glycemia among high-risk patients in North India.
Abstract: Background: Dysglycemia is a major and increasingly prevalent cardiometabolic risk factor worldwide, but is often undiagnosed even in high-risk patients. We evaluated the impact of protocolized screening for dysglycemia on the prevalence of prediabetes and diabetes among patients presenting with ST-segment elevation myocardial infarction (STEMI) in North India. Methods: We conducted a prospective NORIN STEMI registry-based study of patients presenting with STEMI to two government-funded tertiary care medical centers in New Delhi, India, from January to November 2019. Hemoglobin A1c (HbA1c) was collected at presentation as part of the study protocol, irrespective of baseline glycemic status. Results: Among 3,523 participants (median age 55 years), 855 (24%) had known diabetes. In this group, baseline treatment with statins, sodium-glucose cotransporter 2 inhibitors, or glucagon-like peptide-1 receptor agonists was observed in 14%, <1%, and 1% of patients, respectively. For patients without known diabetes, protocolized inpatient screening identified 737 (28%) to have prediabetes (HbA1c 5.7–6.4%) and 339 (13%) to have newly detected diabetes (HbA1c ≥ 6.5%). Patients with prediabetes (49%), newly detected diabetes (53%), and established diabetes (48%) experienced higher rates of post-MI LV dysfunction as compared to euglycemic patients (42%). In-hospital mortality (5.6% for prediabetes, 5.1% for newly detected diabetes, 10.3% for established diabetes, 4.3% for euglycemia) and 30-day mortality (8.1%, 7.6%, 14.4%, 6.6%) were higher in patients with dysglycemia. Compared with euglycemia, prediabetes (adjusted odds ratio (aOR) 1.44 [1.12–1.85]), newly detected diabetes (aOR 1.57 [1.13–2.18]), and established diabetes (aOR 1.51 [1.19–1.94]) were independently associated with higher odds of composite 30-day all-cause mortality or readmission. Conclusions: Among patients presenting with STEMI in North India, protocolized HbA1c screening doubled the proportion of patients with known dysglycemia. Dysglycemia was associated with worse clinical outcomes at 30 days, and use of established pharmacotherapeutic risk-reduction strategies among patients with known diabetes was rare, highlighting missed opportunities for screening and management of dysglycemia among high-risk patients in North India.
TL;DR: In this article , the authors investigated whether females with rheumatic mitral valve disease are more predisposed to develop pulmonary hypertension compared to males, and found that the presence of ET-1 and ETA gene polymorphisms were independent predictors of development of pulmonary arterial hypertension in females.
Abstract: The female gender is a risk factor for idiopathic pulmonary arterial hypertension. However, it is unknown whether females with rheumatic mitral valve disease are more predisposed to develop pulmonary hypertension compared to males.We aimed to investigate whether there was a difference in genotypic distribution of endothelin-1 (ET-1) and endothelin receptor A (ETA) genes between female and male patients of pulmonary hypertension associated with rheumatic mitral valve disease (PH-MVD).We compared prevalence of ET-1 gene (Lys198Asn) and ETA gene (His323His) polymorphisms according to gender in 123 PH-MVD subjects and 123 healthy controls.The presence of mutant Asn/Asn and either mutant Asn/Asn or heterozygous Lys/Asn genotypes of Lys198Asn polymorphism when compared to Lys/Lys in females showed significant association with higher risk (odds ratio [OR] 4.5; p =0.007 and OR 2.39; p =0.02, respectively). The presence of heterozygous C/T and either mutant T/T or heterozygous C/T genotypes of His323His polymorphism when compared to wild C/C genotype in females showed a significant association with higher risk (OR 1.96; p =0.047 and OR 2.26; p =0.01, respectively). No significant difference was seen in genotypic frequencies in males between PH-MVD subjects and controls. Logistic regression analysis showed that mutant genotype Asn/Asn (p =0.007) and heterozygous genotype Lys/Asn of Lys198Asn polymorphism (p =0.018) were independent predictors of development of PH in females.ET-1 and ETA gene polymorphisms were more prevalent in females than males in PH-MVD signifying that females with rheumatic heart disease may be more susceptible to develop PH.
TL;DR: The need for close follow-up of COVID-19 recovered subjects in order to determine the long-term cardiovascular outcomes is highlighted, as patients with severe disease during index admission had far lower LV and RVGLS as compared to mild and moderate cases.
Abstract: Abstract Funding Acknowledgements Type of funding sources: None. Introduction Myocardial injury during acute COVID-19 infection is well characterised however, its persistence during recovery is unclear. Purpose We assessed left ventricle (LV) global longitudinal strain (GLS) and right ventricular (RV) free wall longitudinal strain and RV global longitudinal strain (RV-GLS) using speckle tracking echocardiography (STE) in COVID-19 recovered patients (30-45 days post recovery) and studied its correlation with various parameters. Methods Of the 245 subjects screened, a total of 53 subjects recovered from COVID-19 infection and normal LV ejection fraction were enrolled. Routine blood investigations, inflammatory markers (on admission) and comprehensive echocardiography including STE were done for all. Results All the 53 subjects were symptomatic during COVID-19 illness and were categorized as mild: 27 (50.9%), moderate: 20 (37.7%) and severe: 6 (11.4%) COVID-19 illness. Reduced LV GLS was reported in 22 (41.5%), reduced RV-GLS in 23 (43.4%) and reduced RVFWS in 22 (41.5%) patients respectively. LVGLS was significantly lower in patients recovered from severe illness (mild: -20.3 ± 1.7%; moderate: -15.3 ± 3.4%; severe: -10.7 ± 5.1%; P < 0.0001). Similarly, RVGLS (mild: -21.8 ± 2.8%; moderate: -16.8 ± 4.8%; severe: -9.7 ± 4.6%; P < 0.0001) and RVFWS (mild: -23.0 ± 4.1%; moderate: -18.1 ± 5.5%; severe: -9.3 ± 4.4%; P < 0.0001) were significantly lower in subjects with severe COVID-19. Subjects with reduced LVGLS as well as RVGLS and RVFWS had significantly higher interleukin-6, C-reactive protein, lactate dehydrogenase and serum ferritin levels during index admission. Conclusions Subclinical LV and RV dysfunction was seen in majority of COVID-19 recovered patients. Patients with severe disease during index admission had far lower LV and RVGLS as compared to mild and moderate cases. Our study highlights the need for close follow-up of COVID-19 recovered subjects in order to determine the long-term cardiovascular outcomes.
24 Jun 2022
TL;DR: XMD as discussed by the authors exploits thread-level profiling power of the CPU-core telemetry, and the global profiling power for non-core channels, to achieve significantly better detection performance than currently used Hardware Performance Counter (HPC) based detectors.
Abstract: Hardware-based Malware Detectors (HMDs) have shown promise in detecting malicious workloads. However, the current HMDs focus solely on the CPU core of a System-on-Chip (SoC) and, therefore, do not exploit the full potential of the hardware telemetry. In this paper, we propose XMD, an HMD that uses an expansive set of telemetry channels extracted from the different subsystems of SoC. XMD exploits the thread-level profiling power of the CPU-core telemetry, and the global profiling power of non-core telemetry channels, to achieve significantly better detection performance than currently used Hardware Performance Counter (HPC) based detectors. We leverage the concept of manifold hypothesis to analytically prove the performance gains observed in XMD. We train and evaluate XMD using hardware telemetries collected from 904 benign applications and 1205 malware samples on a commodity Android Operating System (OS)-based mobile device. XMD improves over currently used HPC-based detectors by 32.91% for the in-distribution test data. XMD achieves the best detection performance of 86.54% with a false positive rate of 2.9%, compared to the detection rate of 80\%, offered by the best performing software-based Anti-Virus(AV) on VirusTotal, on the same set of malware samples.
TL;DR: In this paper , a low-overhead energy harvesting and delivery system (EHDS) with pulse frequency modulated (PFM) integrated voltage regulator (IVR) power conversion and self-tuned maximization of system output power is presented.
Abstract: This article presents a low-overhead energy harvesting and delivery system (EHDS) with pulse frequency modulated (PFM) integrated voltage regulator (IVR) power conversion and self-tuned maximization of system output power. A novel load-inclusive time-based maximum power point tracking (LI-TB-MPPT) is developed to provide centralized tuning of PFM-IVR operation based on both source capabilities and load demand on-the-fly, and a configurable fractional sample and hold circuit provides adaptive harvesting window control. The proposed EHDS enables robust harvesting while relieving the use of high passives, with over two orders of magnitude reduction, at the cost of only slight decrease in end-to-end efficiency compared to prior works. Furthermore, a low-overhead wake-up assist circuit utilizes cold-configuration of harvesting sources for efficient and accelerated cold-start. The proposed EHDS is demonstrated in a 65 nm CMOS process with commercial photovoltaic energy harvesting modules. Using only 1.2 and 1 $\mu$H of passives, measured results show a peak 74.9% end-to-end efficiency (simulated up to 85% at 47$\mu$H) and a fast startup time of 3.8 ms. Up to 15% increase in conversion efficiency against load and input voltage variations is achieved with LI-TB-MPPT. The results demonstrate a compact solution for self-sustained cost-restricted stand-alone systems.
TL;DR: This work develops an unsupervised Hebbian learning method in a set-to-cluster module to efficiently create a variable number of the clusters, and it features lower inference time complexity than conventional k-means clustering.
Abstract: —Tactics in StarCraft II are closely related to group behavior of the game agents. In other words, human players in the game often group spatially near agents into a team and control the team to defeat opponents. In this light, clustering the agents in StarCraft II has been studied for various purposes such as the efficient control of the agents in multi-agent reinforcement learning and game analytic tools for the game users. However, these works do not aim to learn and predict dynamics of the clusters, limiting the applications to currently observed game status. In this paper, we present a hybrid AI model that couples unsupervised and self-supervised learning to forecast evolution of the clusters in StarCraft II. We develop an unsupervised Hebbian learning method in a set-to-cluster module to efficiently create a variable number of the clusters, and it also features lower inference time complexity than conventional k-means clustering. For the prediction task, a long short-term memory based prediction module is designed to recursively forecast state vectors generated by the set-to-cluster module. We observe the proposed model successfully predicts complex evolution of the clusters with regard to cluster centroids and their radii. Impact Statement —While there have been many efforts for AI models in StarCraft II to efficiently control or display the game agents by clustering, how the clusters evolve across both space and time is rarely studied. As dynamics in the game is difficult to mathematically design, forecasting evolution of such complex dynamical system by data-driven approach is an important task in recent AI research. In this work, we present a hybrid AI model to forecast evolution of the agents in a cluster level. The proposed model efficiently predicts the complex configurations of the clusters such as where new clusters will appear and how large they will be. In the future, we expect that our model can be also applied to other multi-agent systems where agents often behave as a team.
TL;DR: The final architecture obtained from Model Uncertainty-aware Differentiable ARchiTecture Search shows higher robustness to noise at the input image and model parameters compared to the Architecture obtained from existing DARTS methods.
Abstract: We present a Model Uncertainty-aware Differentiable ARchiTecture Search ( $\mu $ DARTS) that optimizes neural networks to simultaneously achieve high accuracy and low uncertainty. We introduce concrete dropout within DARTS cells and include a Monte-Carlo regularizer within the training loss to optimize the concrete dropout probabilities. A predictive variance term is introduced in the validation loss to enable searching for architecture with minimal model uncertainty. The experiments on CIFAR10, CIFAR100, SVHN, and ImageNet verify the effectiveness of $\mu $ DARTS in improving accuracy and reducing uncertainty compared to existing DARTS methods. Moreover, the final architecture obtained from $\mu $ DARTS shows higher robustness to noise at the input image and model parameters compared to the architecture obtained from existing DARTS methods.
19 Aug 2022
TL;DR: In this paper , a hybrid AI model that couples unsupervised and self-supervised learning to forecast evolution of the clusters in StarCraft II is presented, which can be used for efficient control of the agents in multi-agent reinforcement learning and game analytic tools for the game users.
Abstract: Large multi-agent systems such as real-time strategy games are often driven by collective behavior of agents. For example, in StarCraft II, human players group spatially near agents into a team and control the team to defeat opponents. In this light, clustering the agents in the game has been used for various purposes such as the efficient control of the agents in multi-agent reinforcement learning and game analytic tools for the game users. However, despite the useful information provided by clustering, learning the dynamics of multi-agent systems at a cluster level has been rarely studied yet. In this paper, we present a hybrid AI model that couples unsupervised and self-supervised learning to forecast evolution of the clusters in StarCraft II. We develop an unsupervised Hebbian learning method in a set-to-cluster module to efficiently create a variable number of the clusters with lower inference time complexity than K-means clustering. Also, a long short-term memory based prediction module is designed to recursively forecast state vectors generated by the set-to-cluster module to define cluster configuration. We experimentally demonstrate the proposed model successfully predicts complex movement of the clusters in the game.
28 Oct 2022
TL;DR: In this article , a CNN-LSTM model is proposed to forecast the state of a particular agent in a large self-organizing multi-agent system without the reconstruction, where the model comprises a CNN encoder to represent the system in a low-dimensional vector, a LSTM module to learn agent dynamics in the vector space, and a MLP decoder to predict the future state of an agent.
Abstract: Large multi-agent systems are often driven by locally defined agent interactions, which is referred to as self-organization. Our primary objective is to determine when the propagation of such local interactions will reach a specific agent of interest. Although conventional approaches that reconstruct all agent states can be used, they may entail unnecessary computational costs. In this paper, we investigate a CNN-LSTM model to forecast the state of a particular agent in a large self-organizing multi-agent system without the reconstruction. The proposed model comprises a CNN encoder to represent the system in a low-dimensional vector, a LSTM module to learn agent dynamics in the vector space, and a MLP decoder to predict the future state of an agent. As an example, we consider a forest fire model where we aim to predict when a particular tree agent will start burning. We compare the proposed model with reconstruction-based approaches such as CNN-LSTM and ConvLSTM. The proposed model exhibits similar or slightly worse AUC but significantly reduces computational costs such as activation than ConvLSTM. Moreover, it achieves higher AUC with less computation than the recontruction-based CNN-LSTM.
TL;DR: In this paper , the role of MMPs and their genetic polymorphisms in young (<50 years) ST-segment elevation myocardial infarction (STEMI) patients was investigated.
Abstract: Genetic polymorphism in MMPs are associated with multiple adverse CV events. There is little evidence regarding role of MMPs and their genetic polymorphisms in young (<50 years) ST-segment elevation myocardial infarction (STEMI) patients.This study included 100 young (18-50 years) STEMI patients and 100 healthy controls. Serum levels of MMP-3, MMP-9 and TIMP were estimated for both patients as well as controls. Additionally, genetic polymorphisms in the MMP-9 gene (-1562 C/T and R279Q) & MMP-3 gene (5A/6A-1612) was evaluated. All these patients were followed up for one year and major adverse cardiac events (MACE) were determined.Serum levels of MMP-3 (128.16 ± 115.81 vs 102.3 ± 57.28 ng/mL; P = 0.04), MMP-9 (469.63 ± 238.4 vs 188.88 ± 94.08 pg/mL; P < 0.0001) and TIMP (5.84 ± 1.93 vs 2.28 ± 1.42 ng/mL; P < 0.0001) were significantly higher in patients as compared to controls. Additionally, patients with genetic polymorphisms in the MMP genes (5A/5A, 6A/6A and the AG genotypes) had an increased risk of STEMI. Patients with MACE had significantly higher levels of MMP-9 (581.73 ± 260.93 vs 438.01 ± 223.38 pg/mL; P = 0.012). A cutoff value of 375.5 pg/mL of MMP-9 was best able to discriminate patients with STEMI and MACE with sensitivity of 77.3% and specificity of 57%.Novel biomarkers such as MMP-3, MMP-9 and TIMP and their genetic polymorphism are associated with the susceptibility for STEMI in young individuals. Higher MMP-9 levels in STEMI patients with MACE suggests its potential role in predicting cardiac remodeling and left ventricular dysfunction.
DOI•
19 Jun 2022
TL;DR: A hybrid radio frequency machine learning (RFML) model that couples Short-time Fourier Transform with Convolutional Neural Network (STFT-CNN) for Automatic Modulation Classification (AMC) and the simulation shows higher average accuracy than a time-domain CNN accelerator with 32-bit floating point operation.
Abstract: We present a hybrid radio frequency machine learning (RFML) model that couples Short-time Fourier Transform with Convolutional Neural Network (STFT-CNN) for Automatic Modulation Classification (AMC). The simulation on RadioML2016.10a show 77.9% average accuracy for 0dB or higher Signal to Noise ratio with 16-bit fixed-point operation. An on-chip accelerator for STFT-CNN, designed and synthesized in 28nm CMOS, shows 7× lower power, 2× lower processing time, and 7.5× lower memory than a time-domain CNN accelerator with 32-bit floating point operation.
TL;DR: A novel ranking method to rank videos based on the degree of global camera motion and a novel input dependent weighted averaging strategy for fusing local and global features.
Abstract: —Interactive autonomous applications require ro- bustness of the perception engine to artifacts in unconstrained videos. In this paper, we examine the effect of camera motion on the task of action detection. We develop a novel ranking method to rank videos based on the degree of global camera motion. For the high ranking camera videos we show that the accuracy of action detection is decreased. We propose an action detection pipeline that is robust to the camera motion effect and verify it empirically. Specifically, we do actor feature alignment across frames and couple global scene features with local actor-specific features. We do feature alignment using a novel formulation of the Spatio-temporal Sampling Network (STSN) but with multi-scale offset prediction and refinement using a pyramid structure. We also propose a novel input dependent weighted averaging strategy for fusing local and global features. We show the applicability of our network on our dataset of moving camera videos with high camera motion (MOVE dataset) with a 4.1% increase in frame mAP and 17% increase in video mAP.
TL;DR: In this article , a CNN-LSTM model was proposed to forecast the state of a tree agent in a large multi-agent system, which achieved higher AUC with less computation than a frame-based model and significantly save computational costs such as the activation.
Abstract: In this paper, we study a CNN-LSTM model to forecast the state of a specific agent in a large multi-agent system. The proposed model consists of a CNN encoder to represent the system into a low-dimensional vector, a LSTM module to learn the agent dynamics in the vector space, and a MLP decoder to predict the future state of an agent. A forest fire model is considered as an example where we need to predict when a specific tree agent will be burning. We observe that the proposed model achieves higher AUC with less computation than a frame-based model and significantly saves computational costs such as the activation than ConvLSTM.
TL;DR: In this article , the authors investigated the performance of three different OAT methods: (1) model-based methods (i.e., classical ray-based OAT using a linearized forward model), (2) data-driven methods (such as deep learning) to directly learn the inverse model, and (3) a hybrid solution which combines deep learning of the forward model followed by a standard recursive optimization to estimate SSPs.
Abstract: Underwater sound propagation is primarily driven by a non-linear forward model relating variability of the ocean sound speed profile (SSP) to the acoustic observations (e.g., eigenray arrival times). Ocean acoustic tomography (OAT) methods aim at reconstructing SSPs variations (with respect to a reference environment) from changes of the acoustic measurements between multiple source-receiver pairs. We investigated the performance of three different OAT methods: (1) model-based methods (i.e., classical ray-based OAT using a linearized forward model), (2) data-driven methods (such as deep learning) to directly learn the inverse model, and (3) a hybrid solution (i.e., the Neural Adjoint-NA- method) which combines deep learning of the forward model followed by a standard recursive optimization to estimate SSPs. Additionally, synthetic SSPs were generated to augment the variability of the training set. We tested these methods with modeled ray arrivals calculated for a downward refracting environment with mild fluctuations of the thermocline using idealized towed and fixed source configurations. Results indicate that merging data-driven and model-based methods can benefit OAT predictions depending on the selected sensing configurations and actual ray coverage of the water column. But ultimately, the robustness of the OAT predictions depends on the actual dynamics of the SSP variations.