Searcharxiv⌕ Search

arXiv subjects

Akash Kumar

Publications and source records attributed to Akash Kumar.

At least 73 records · Page 4Linked to original sources

Stable Mean Teacher for Semi-supervised Video Action Detection

In this work, we focus on semi-supervised learning for video action detection. Video action detection requires spatiotemporal localization in addition to classification, and a limited amount of labels makes the model prone to unreliable predictions. We present Stable Mean Teacher, a simple end-to-end teacher-based framework that benefits from improved and temporally consistent pseudo labels. It relies on a novel Error Recovery (EoR) module, which learns from students' mistakes on labeled samples and transfers this knowledge to the teacher to improve pseudo labels for unlabeled samples. Moreover, existing spatiotemporal losses do not take temporal coherency into account and are prone to temporal inconsistencies. To address this, we present Difference of Pixels (DoP), a simple and novel constraint focused on temporal consistency, leading to coherent temporal detections. We evaluate our approach on four different spatiotemporal detection benchmarks: UCF101-24, JHMDB21, AVA, and YouTube-VOS. Our approach outperforms the supervised baselines for action detection by an average margin of 23.5% on UCF101-24, 16% on JHMDB21, and 3.3% on AVA. Using merely 10% and 20% of data, it provides competitive performance compared to the supervised baseline trained on 100% annotations on UCF101-24 and JHMDB21, respectively. We further evaluate its effectiveness on AVA for scaling to large-scale datasets and YouTube-VOS for video object segmentation, demonstrating its generalization capability to other tasks in the video domain. Code and models are publicly available.

cs.CV↗

Mutual synchronization in spin torque and spin Hall nano-oscillators

This chapter reviews the state of the art in mutually synchronized spin-torque and spin Hall nano-oscillator (STNO and SHNO) arrays. After briefly introducing the underlying physics, we discuss different nano-oscillator implementations and their functional properties with respect to frequency range, output power, phase noise, and modulation rates. We then introduce the concepts and the theory of mutual synchronization and discuss the possible coupling mechanisms in spintronic nano-oscillators, such as dipolar, electrical, and spin-wave coupling. We review the experimental literature on mutually synchronized STNOs and SHNOs in one- and two-dimensional arrays and discuss ways to increase the number of mutually synchronized nano-oscillators. Finally, the potential for applications ranging from microwave signal sources/detectors and ultrafast spectrum analyzers to neuromorphic computing elements and Ising machines is discussed together with the specific electronic circuitry that has been designed so far to harness this potential.

cond-mat.mes-hall↗

Effect of 2009 major SSW event on the mesospheric CO2 cooling

Carbon dioxide (CO2), an important trace species that is gradually increasing in the atmosphere due to anthropogenic activities, causes enhanced warming in the lower atmosphere. The increased concentration of CO2 in the upper atmosphere results in enhanced radiative cooling rates leading to the contraction of the upper atmosphere. Due to its long lifetime and large vertical gradient, CO2 concentration is also influenced by large dynamic events. We report a startling case of variability in CO2 density and its infrared radiative cooling rates in the mesosphere and lower thermospher during a major sudden stratospheric warming (SSW) event. A counter-intuitive connection between CO2 density and resulting CO2 radiative cooling has been observed during the 2009 major SSW event. The behaviour of CO2 cooling rates during such a dramatic events draw attention to our current understanding of CO2 infrared cooling variation and its connection to changes in CO2 concentration. The significance of temperature and atomic oxygen variability in the observed cooling patterns despite changes in CO2 concentration, is also highlighted.

physics.space-ph↗

Learning Smooth Distance Functions via Queries

In this work, we investigate the problem of learning distance functions within the query-based learning framework, where a learner is able to pose triplet queries of the form: ``Is $x_i$ closer to $x_j$ or $x_k$?'' We establish formal guarantees on the query complexity required to learn smooth, but otherwise general, distance functions under two notions of approximation: $ω$-additive approximation and $(1 + ω)$-multiplicative approximation. For the additive approximation, we propose a global method whose query complexity is quadratic in the size of a finite cover of the sample space. For the (stronger) multiplicative approximation, we introduce a method that combines global and local approaches, utilizing multiple Mahalanobis distance functions to capture local geometry. This method has a query complexity that scales quadratically with both the size of the cover and the ambient space dimension of the sample space.

cs.LG↗

Mirror Descent on Reproducing Kernel Banach Spaces

Recent advances in machine learning have led to increased interest in reproducing kernel Banach spaces (RKBS) as a more general framework that extends beyond reproducing kernel Hilbert spaces (RKHS). These works have resulted in the formulation of representer theorems under several regularized learning schemes. However, little is known about an optimization method that encompasses these results in this setting. This paper addresses a learning problem on Banach spaces endowed with a reproducing kernel, focusing on efficient optimization within RKBS. To tackle this challenge, we propose an algorithm based on mirror descent (MDA). Our approach involves an iterative method that employs gradient steps in the dual space of the Banach space using the reproducing kernel. We analyze the convergence properties of our algorithm under various assumptions and establish two types of results: first, we identify conditions under which a linear convergence rate is achievable, akin to optimization in the Euclidean setting, and provide a proof of the linear rate; second, we demonstrate a standard convergence rate in a constrained setting. Moreover, to instantiate this algorithm in practice, we introduce a novel family of RKBSs with $p$-norm ($p \neq 2$), characterized by both an explicit dual map and a kernel.

cs.LG↗

Ultra-fast spin Hall nano-oscillator based microwave spectral analysis

Ultra-fast spectrum analysis concept based on rapidly tuned spintronic nano-oscillators has been under development for the last few years and has already demonstrated promising results. Here, we demonstrate an ultra-fast microwave spectrum analyzer based on a chain of five mutually synchronized nano-constriction spin Hall nano-oscillators (SHNOs). As mutual synchronization affords the chain a much improved signal quality, with linewidths well below 1 MHz at close to a 10 GHz operating frequency, we observe an order of magnitude better frequency resolution bandwidth compared to previously reported spectral analysis based on single magnetic tunnel junction based spin torque nano-oscillators. The high-frequency operation and ability to synchronize long SHNO chains and large arrays make SHNOs ideal candidates for ultra-fast microwave spectral analysis.

cond-mat.mes-hall↗

QoS-Nets: Adaptive Approximate Neural Network Inference

In order to vary the arithmetic resource consumption of neural network applications at runtime, this work proposes the flexible reuse of approximate multipliers for neural network layer computations. We introduce a search algorithm that chooses an appropriate subset of approximate multipliers of a user-defined size from a larger search space and enables retraining to maximize task performance. Unlike previous work, our approach can output more than a single, static assignment of approximate multiplier instances to layers. These different operating points allow a system to gradually adapt its Quality of Service (QoS) to changing environmental conditions by increasing or decreasing its accuracy and resource consumption. QoS-Nets achieves this by reassigning the selected approximate multiplier instances to layers at runtime. To combine multiple operating points with the use of retraining, we propose a fine-tuning scheme that shares the majority of parameters between operating points, with only a small amount of additional parameters required per operating point. In our evaluation on MobileNetV2, QoS-Nets is used to select four approximate multiplier instances for three different operating points. These operating points result in power savings for multiplications between 15.3% and 42.8% at a Top-5 accuracy loss between 0.3 and 2.33 percentage points. Through our fine-tuning scheme, all three operating points only increase the model's parameter count by only 2.75%.

cs.LG↗

Seasonal variation in nighttime NO radiative cooling as observed by TIMED/SABER in lower thermosphere during solar maximum and solar minimum

Both composition and temperature play a crucial role in determining the NO radiative cooling in lower thermosphere as observed by TIMED/SABER. In this work, we present a detailed investigation of seasonal variation in thermospheric NO radiative cooling. We have carried forward the investigation of \cite{li2018} regarding the variations in local nighttime peak NO radiative cooling and its altitude during solar maximum and solar minimum conditions. By analyzing latitudinal changes over quiet times for each month in year 2018, it is evident that both the investigative parameters exhibit summer-winter variability. The qualitative contribution of different species (i.e., NO, and O), and temperatures in determining the vertical profile of NO radiative cooling for different latitudes is investigated by utilizing the NRLMSISE-00 estimated parameters, and SNOE observed NO density. The temperature, NO density, meridional wind, and associated compositional variations due to asymmetrical solar heating in both the hemispheres during solar minimum conditions seem to be the dominating factor in controlling the NO radiative cooling during different seasons. The altitudes at which maximum cooling by NO occurs exhibits an inverse correlation with the amount of radiative cooling. The region of enhanced NO densities (polar and summer hemispheric low-mid latitude regions) have larger NO radiative cooling with lower peak altitudes in comparison to other regions (equatorial to winter hemispheric low-mid latitude regions), where NO radiative cooling is low with higher peak altitude values.

physics.space-ph↗

Adsorption of Mo and O at S-vacancy on ReS2 surface of ReS2/MoTe2 vdW heterointerface

Applications like high density information storage, neuromorphic computing, nanophotonics, etc. require ultra-thin electronic devices which can be controlled with applied electric field. Of late, atomically thin two-dimensional (2D) materials and van der Waals (vdW) heterointerface of those have emerged as suitable candidates for such ultra-low power nanoelectric devices. In this work, employing density functional theory (DFT), the monolayer ReS2 / monolayer MoTe2 vdW heterostructure with Sulphur vacancy is studied to examine various ground state electronic properties. Changes in effective band gap owing to defect-induced states and modulation of the energy gap value with Molybdenum (Mo) and Oxygen (O) adsorption at the defect site are examined. Since two-dimensional (2D) material based nanoscaled devices exhibit promising switching between non-conducting and conducting states, determining the role of defect-induced states and the adsorption of atoms/molecules on surfaces is crucial. Here, a detailed theoretical study to determine surface properties and relative energetic stability of the vdW heterostructures is carried out. The charge re-distribution between the constituent layers is also analyzed by obtaining Electron Difference Density (EDD) for different heterointerfaces. Nonetheless, the efficacy of switching between non-conducting and conducting states is assessed based on adsorption energy of adatoms binding at the defect site.

cond-mat.mtrl-sci↗

Ultra-low-current-density single-layer magnetic Weyl semimetal spin Hall nano-oscillators

Topological quantum materials can exhibit unconventional surface states and anomalous transport properties. Still, their applications in spintronic devices are restricted as they require the growth of high-quality thin films with bulk-like properties. Here, we study 10--30 nm thick epitaxial ferromagnetic Co$_{\rm 2}$MnGa films with high structural order and very high values of the anomalous Hall conductivity, $σ_{\rm xy}=1.35\times10^{5}$ $Ω^{-1} m^{-1}$ and the anomalous Hall angle, $θ_{\rm H}=15.8\%$, both comparable to bulk values. We observe a dramatic crystalline orientation dependence of the Gilbert damping constant of a factor of two and a giant intrinsic spin Hall conductivity, $\mathit{σ_{\rm SHC}}=(6.08\pm 0.02)\times 10^{5}$ ($\hbar/2e$) $Ω^{-1} m^{-1}$, an order of magnitude higher than literature values of multilayer Co$_{\rm 2}$MnGa stacks [1-3] and single-layer Ni, Co, Fe [4], and Ni$_{\rm 80}$Fe$_{\rm 20}$~[4,5]. As a consequence, spin-orbit-torque driven auto-oscillations of a 30 nm thick magnetic film are observed for the first time, at an ultralow threshold current density of $J_{th}=6.2\times10^{11}$ $Am^{-2}$. Theoretical calculations of the intrinsic spin Hall conductivity, originating from a strong Berry curvature, corroborate the results and yield values comparable to the experiment. Our results open up for the design of spintronic devices based on single layers of magnetic topological quantum materials.

cond-mat.mes-hall↗

Spin Hall Nano-Oscillator Empirical Electrical Model for Optimal On-chip Detector Design

As nascent nonlinear oscillators, nano-constriction spin Hall nano-oscillators (SHNOs) represent a promising potential for integration into more complicated systems such as neural networks, magnetic field sensors, and radio frequency (RF) signal classification, their tunable high-frequency operating regime, easy synchronization, and CMOS compatibility can streamline the process. To implement SHNOs in any of these networks, the electrical features of a single device are needed before designing the signal detection CMOS circuitry. This study centers on presenting an empirical electrical model of the SHNO based on a comprehensive characterization of the output impedance of a single SHNO, and its available output power in the range of 2-10 GHz at various bias currents.

cond-mat.mes-hall↗

Semi-supervised Active Learning for Video Action Detection

In this work, we focus on label efficient learning for video action detection. We develop a novel semi-supervised active learning approach which utilizes both labeled as well as unlabeled data along with informative sample selection for action detection. Video action detection requires spatio-temporal localization along with classification, which poses several challenges for both active learning informative sample selection as well as semi-supervised learning pseudo label generation. First, we propose NoiseAug, a simple augmentation strategy which effectively selects informative samples for video action detection. Next, we propose fft-attention, a novel technique based on high-pass filtering which enables effective utilization of pseudo label for SSL in video action detection by emphasizing on relevant activity region within a video. We evaluate the proposed approach on three different benchmark datasets, UCF-101-24, JHMDB-21, and Youtube-VOS. First, we demonstrate its effectiveness on video action detection where the proposed approach outperforms prior works in semi-supervised and weakly-supervised learning along with several baseline approaches in both UCF101-24 and JHMDB-21. Next, we also show its effectiveness on Youtube-VOS for video object segmentation demonstrating its generalization capability for other dense prediction tasks in videos. The code and models is publicly available at: \url{https://github.com/AKASH2907/semi-sup-active-learning}.

cs.CV↗

Spintronic devices as next-generation computation accelerators

The ever increasing demand for computational power combined with the predicted plateau for the miniaturization of existing silicon-based technologies has made the search for low power alternatives an industrial and scientifically engaging problem. In this work, we explore spintronics-based Ising machines as hardware computation accelerators. We start by presenting the physical platforms on which this emerging field is being developed, the different control schemes and the type of algorithms and problems on which these machines outperform conventional computers. We then benchmark these technologies and provide an outlook for future developments and use-cases that can help them get a running start for integration into the next generation of computing devices.

cond-mat.mes-hall↗

Efficient Post-Training Augmentation for Adaptive Inference in Heterogeneous and Distributed IoT Environments

Early Exit Neural Networks (EENNs) present a solution to enhance the efficiency of neural network deployments. However, creating EENNs is challenging and requires specialized domain knowledge, due to the large amount of additional design choices. To address this issue, we propose an automated augmentation flow that focuses on converting an existing model into an EENN. It performs all required design decisions for the deployment to heterogeneous or distributed hardware targets: Our framework constructs the EENN architecture, maps its subgraphs to the hardware targets, and configures its decision mechanism. To the best of our knowledge, it is the first framework that is able to perform all of these steps. We evaluated our approach on a collection of Internet-of-Things and standard image classification use cases. For a speech command detection task, our solution was able to reduce the mean operations per inference by 59.67%. For an ECG classification task, it was able to terminate all samples early, reducing the mean inference energy by 74.9% and computations by 78.3%. On CIFAR-10, our solution was able to achieve up to a 58.75% reduction in computations. The search on a ResNet-152 base model for CIFAR-10 took less than nine hours on a laptop CPU. Our proposed approach enables the creation of EENN optimized for IoT environments and can reduce the inference cost of Deep Learning applications on embedded and fog platforms, while also significantly reducing the search cost - making it more accessible for scientists and engineers in industry and research. The low search cost improves the accessibility of EENNs, with the potential to improve the efficiency of neural networks in a wide range of practical applications.

cs.LG↗

Temporal Decisions: Leveraging Temporal Correlation for Efficient Decisions in Early Exit Neural Networks

Deep Learning is becoming increasingly relevant in Embedded and Internet-of-things applications. However, deploying models on embedded devices poses a challenge due to their resource limitations. This can impact the model's inference accuracy and latency. One potential solution are Early Exit Neural Networks, which adjust model depth dynamically through additional classifiers attached between their hidden layers. However, the real-time termination decision mechanism is critical for the system's efficiency, latency, and sustained accuracy. This paper introduces Difference Detection and Temporal Patience as decision mechanisms for Early Exit Neural Networks. They leverage the temporal correlation present in sensor data streams to efficiently terminate the inference. We evaluate their effectiveness in health monitoring, image classification, and wake-word detection tasks. Our novel contributions were able to reduce the computational footprint compared to established decision mechanisms significantly while maintaining higher accuracy scores. We achieved a reduction of mean operations per inference by up to 80% while maintaining accuracy levels within 5% of the original model. These findings highlight the importance of considering temporal correlation in sensor data to improve the termination decision.

cs.LG↗

Spin wave-driven variable-phase mutual synchronization in spin Hall nano-oscillators

Spin-orbit torque can drive auto-oscillations of propagating spin wave (PSW) modes in nano-constriction spin Hall nano-oscillators (SHNOs). These modes allow both long-range coupling and the potential of controlling its phase -- critical aspect for nano-magnonics, spin wave logic, and Ising machines. Here, we demonstrate PSW-driven variable-phase coupling between two nano-constriction SHNOs and study how their separation and the PSW wave vector impact their mutual synchronization. In addition to ordinary in-phase mutual synchronization, we observe, using both electrical measurements and phase-resolved $μ-$Brillouin Light Scattering microscopy, mutual synchronization with a phase that can be tuned from 0 to $π$ using the drive current or the applied field. Micromagnetic simulations corroborate the experiments and visualize how the PSW patterns in the bridge connecting the two nano-constrictions govern the coupling. These results advance the capabilities of mutually synchronized SHNOs and open up new possibilities for applications in spin wave logic, unconventional computing, and Ising Machines.

cond-mat.mes-hall↗

Large Non-Volatile Frequency Tuning of Spin Hall Nano-Oscillators using Circular Memristive Nano-Gates

Spin Hall nano oscillators (SHNOs) are promising candidates for neuromorphic computing due to their miniaturized dimensions, non-linearity, fast dynamics, and ability to synchronize in long chains and arrays. However, tuning the individual SHNOs in large chains/arrays, which is key to implementing synaptic control, has remained a challenge. Here, we demonstrate circular memristive nano-gates, both precisely aligned and shifted with respect to nano-constriction SHNOs of W/CoFeB/HfOx, with increased quality of the device tunability. Gating at the exact center of the nano-constriction region is found to cause irreversible degradation to the oxide layer, resulting in a permanent frequency shift of the auto-oscillating modes. As a remedy, gates shifted outside of the immediate nano-constriction region can tune the frequency dramatically (>200 MHz) without causing any permanent change to the constriction region. Circular memristive nano-gates can, therefore, be used in SHNO chains/arrays to manipulate the synchronization states precisely over large networks of oscillators.

physics.app-ph↗

A Large-Scale Analysis on Self-Supervised Video Representation Learning

Self-supervised learning is an effective way for label-free model pre-training, especially in the video domain where labeling is expensive. Existing self-supervised works in the video domain use varying experimental setups to demonstrate their effectiveness and comparison across approaches becomes challenging with no standard benchmark. In this work, we first provide a benchmark that enables a comparison of existing approaches on the same ground. Next, we study five different aspects of self-supervised learning important for videos; 1) dataset size, 2) complexity, 3) data distribution, 4) data noise, and, 5)feature analysis. To facilitate this study, we focus on seven different methods along with seven different network architectures and perform an extensive set of experiments on 5 different datasets with an evaluation of two different downstream tasks. We present several interesting insights from this study which span across different properties of pretraining and target datasets, pretext-tasks, and model architectures among others. We further put some of these insights to the real test and propose an approach that requires a limited amount of training data and outperforms existing state-of-the-art approaches which use 10x pretraining data. We believe this work will pave the way for researchers to a better understanding of self-supervised pretext tasks in video representation learning.

cs.CV↗