Searcharxiv⌕ Search

arXiv subjects

Akash Kumar

Publications and source records attributed to Akash Kumar.

At least 91 records · Page 5Linked to original sources

AxOMaP: Designing FPGA-based Approximate Arithmetic Operators using Mathematical Programming

With the increasing application of machine learning (ML) algorithms in embedded systems, there is a rising necessity to design low-cost computer arithmetic for these resource-constrained systems. As a result, emerging models of computation, such as approximate and stochastic computing, that leverage the inherent error-resilience of such algorithms are being actively explored for implementing ML inference on resource-constrained systems. Approximate computing (AxC) aims to provide disproportionate gains in the power, performance, and area (PPA) of an application by allowing some level of reduction in its behavioral accuracy (BEHAV). Using approximate operators (AxOs) for computer arithmetic forms one of the more prevalent methods of implementing AxC. AxOs provide the additional scope for finer granularity of optimization, compared to only precision scaling of computer arithmetic. To this end, designing platform-specific and cost-efficient approximate operators forms an important research goal. Recently, multiple works have reported using AI/ML-based approaches for synthesizing novel FPGA-based AxOs. However, most of such works limit usage of AI/ML to designing ML-based surrogate functions used during iterative optimization processes. To this end, we propose a novel data analysis-driven mathematical programming-based approach to synthesizing approximate operators for FPGAs. Specifically, we formulate mixed integer quadratically constrained programs based on the results of correlation analysis of the characterization data and use the solutions to enable a more directed search approach for evolutionary optimization algorithms. Compared to traditional evolutionary algorithms-based optimization, we report up to 21% improvement in the hypervolume, for joint optimization of PPA and BEHAV, in the design of signed 8-bit multipliers.

cs.AR↗

AxOCS: Scaling FPGA-based Approximate Operators using Configuration Supersampling

The rising usage of AI and ML-based processing across application domains has exacerbated the need for low-cost ML implementation, specifically for resource-constrained embedded systems. To this end, approximate computing, an approach that explores the power, performance, area (PPA), and behavioral accuracy (BEHAV) trade-offs, has emerged as a possible solution for implementing embedded machine learning. Due to the predominance of MAC operations in ML, designing platform-specific approximate arithmetic operators forms one of the major research problems in approximate computing. Recently there has been a rising usage of AI/ML-based design space exploration techniques for implementing approximate operators. However, most of these approaches are limited to using ML-based surrogate functions for predicting the PPA and BEHAV impact of a set of related design decisions. While this approach leverages the regression capabilities of ML methods, it does not exploit the more advanced approaches in ML. To this end, we propose AxOCS, a methodology for designing approximate arithmetic operators through ML-based supersampling. Specifically, we present a method to leverage the correlation of PPA and BEHAV metrics across operators of varying bit-widths for generating larger bit-width operators. The proposed approach involves traversing the relatively smaller design space of smaller bit-width operators and employing its associated Design-PPA-BEHAV relationship to generate initial solutions for metaheuristics-based optimization for larger operators. The experimental evaluation of AxOCS for FPGA-optimized approximate operators shows that the proposed approach significantly improves the quality-resulting hypervolume for multi-objective optimization-of 8x8 signed approximate multipliers.

cs.AR↗

Temporal Patience: Efficient Adaptive Deep Learning for Embedded Radar Data Processing

Radar sensors offer power-efficient solutions for always-on smart devices, but processing the data streams on resource-constrained embedded platforms remains challenging. This paper presents novel techniques that leverage the temporal correlation present in streaming radar data to enhance the efficiency of Early Exit Neural Networks for Deep Learning inference on embedded devices. These networks add additional classifier branches between the architecture's hidden layers that allow for an early termination of the inference if their result is deemed sufficient enough by an at-runtime decision mechanism. Our methods enable more informed decisions on when to terminate the inference, reducing computational costs while maintaining a minimal loss of accuracy. Our results demonstrate that our techniques save up to 26% of operations per inference over a Single Exit Network and 12% over a confidence-based Early Exit version. Our proposed techniques work on commodity hardware and can be combined with traditional optimizations, making them accessible for resource-constrained embedded platforms commonly used in smart devices. Such efficiency gains enable real-time radar data processing on resource-constrained platforms, allowing for new applications in the context of smart homes, Internet-of-Things, and human-computer interaction.

cs.LG↗

On the List Color Function Threshold

The chromatic polynomial of a graph $G$, denoted $P(G,m)$, is equal to the number of proper $m$-colorings of $G$. The list color function of graph $G$, denoted $P_{\ell}(G,m)$, is a list analogue of the chromatic polynomial that has been studied since the early 1990s, primarily through comparisons with the corresponding chromatic polynomial. It is known that for any graph $G$ there is a $k \in \mathbb{N}$ such that $P_\ell(G,m) = P(G,m)$ whenever $m \geq k$. The list color function threshold of $G$, denoted $τ(G)$, is the smallest $k \geq χ(G)$ such that $P_{\ell}(G,m) = P(G,m)$ whenever $m \geq k$. In 2009, Thomassen asked whether there is a universal constant $α$ such that for any graph $G$, $τ(G) \leq χ_{\ell}(G) + α$, where $χ_{\ell}(G)$ is the list chromatic number of $G$. We show that the answer to this question is no by proving that there exists a constant $C$ such that $τ(K_{2,l}) - χ_{\ell}(K_{2,l}) \ge C\sqrt{l}$ for $l \ge 16$.

math.CO↗

Magnetic Proximity induced efficient charge-to-spin conversion in large area PtSe$_{2}$/Ni$_{80}$Fe$_{20}$ heterostructures

As a topological Dirac semimetal with controllable spin-orbit coupling and conductivity, PtSe$_2$, a transition-metal dichalcogenide, is a promising material for several applications from optoelectric to sensors. However, its potential for spintronics applications is yet to be explored. In this work, we demonstrate that PtSe$_{2}$/Ni$_{80}$Fe$_{20}$ heterostructure can generate a large damping-like current-induced spin-orbit torques (SOT), despite the absence of spin-splitting in bulk PtSe$_{2}$. The efficiency of charge-to-spin conversion is found to be $(-0.1 \pm 0.02)$~nm$^{-1}$ in PtSe$_{2}$/Ni$_{80}$Fe$_{20}$, which is three times that of the control sample, Ni$_{80}$Fe$_{20}$/Pt. Our band structure calculations show that the SOT due to the PtSe$_2$ arises from an unexpectedly large spin splitting in the interfacial region of PtSe$_2$ introduced by the proximity magnetic field of the Ni$_{80}$Fe$_{20}$ layer. Our results open up the possibilities of using large-area PtSe$_{2}$ for energy-efficient nanoscale devices by utilizing the proximity-induced SOT.

cond-mat.mtrl-sci↗

Ultra-low current 10 nm spin Hall nano-oscillators

Nano-constriction based spin Hall nano-oscillators (SHNOs) are at the forefront of spintronics research for emerging technological applications such as oscillator-based neuromorphic computing and Ising Machines. However, their miniaturization to the sub-50 nm width regime results in poor scaling of the threshold current. Here, we show that current shunting through the Si substrate is the origin of this problem and study how different seed layers can mitigate it. We find that an ultra-thin Al$_{2}$O$_{3}$ seed layer and SiN (200 nm) coated p-Si substrates provide the best improvement, enabling us to scale down the SHNO width to a truly nanoscopic dimension of 10 nm, operating at threshold currents below 30 $μ$A. In addition, the combination of electrical insulation and high thermal conductivity of the Al$_{2}$O$_{3}$ seed will offer the best conditions for large SHNO arrays, avoiding any significant temperature gradients within the array. Our state-of-the-art ultra-low operational current SHNOs hence pave an energy-efficient route to scale oscillator-based computing to large dynamical neural networks of linear chains or two-dimensional arrays.

cond-mat.mes-hall↗

Interfacial origin of unconventional spin-orbit torque in Py/$γ-$IrMn$_{3}$

Angle-resolved spin-torque ferromagnetic resonance measurements are carried out in heterostructures consisting of Py (Ni$_{81}$Fe$_{19}$) and a noncollinear antiferromagnetic quantum material $γ-$IrMn$_{3}$. The structural characterization reveals that $γ-$IrMn$_{3}$ is polycrystalline in nature. A large exchange bias of 158~Oe is found in Py/$γ-$IrMn$_{3}$ at room temperature, while $γ-$IrMn$_{3}$/Py and Py/Cu/$γ-$IrMn$_{3}$ exhibited no exchange bias. Regardless of the exchange bias and stacking sequence, we observe a substantial unconventional out-of-plane anti-damping torque when $γ-$IrMn$_{3}$ is in direct contact with Py. The magnitude of the out-of-plane spin-orbit torque efficiency is found to be twice as large as the in-plane spin-orbit torque efficiency. The unconventional spin-orbit torque vanishes when a Cu spacer is introduced between Py and $γ-$IrMn$_{3}$, indicating that the unconventional spin-orbit torque in this system originates at the interface. These findings are important for realizing efficient antiferromagnet-based spintronic devices via interfacial engineering.

cond-mat.mtrl-sci↗

Phase noise analysis of mutually synchronized spin Hall nano-oscillators

The reduction of phase noise in electronic systems is of utmost importance in modern communication and signal processing applications and requires an understanding of the underlying physical processes. Here, we systematically study the phase noise in mutually synchronized chains of nano-constriction spin Hall nano-oscillators (SHNOs). We find that longer chains have improved phase noise figures at low offset frequencies (1/f noise), where chains of two and ten mutually synchronized SHNOs have 2.8 and 6.2 dB lower phase noise than single SHNOs. This is close to the theoretical values of 3 and 10 dB, and the deviation is ascribed to process variations between nano-constrictions. However, at higher offset frequencies (thermal noise), the phase noise unexpectedly increases with chain length, which we ascribe to process variations, a higher operating temperature in the long chains at the same drive current and phase delays in the coupling between nano-constrictions.

cond-mat.mes-hall↗

Robust Empirical Risk Minimization with Tolerance

Developing simple, sample-efficient learning algorithms for robust classification is a pressing issue in today's tech-dominated world, and current theoretical techniques requiring exponential sample complexity and complicated improper learning rules fall far from answering the need. In this work we study the fundamental paradigm of (robust) $\textit{empirical risk minimization}$ (RERM), a simple process in which the learner outputs any hypothesis minimizing its training error. RERM famously fails to robustly learn VC classes (Montasser et al., 2019a), a bound we show extends even to `nice' settings such as (bounded) halfspaces. As such, we study a recent relaxation of the robust model called $\textit{tolerant}$ robust learning (Ashtiani et al., 2022) where the output classifier is compared to the best achievable error over slightly larger perturbation sets. We show that under geometric niceness conditions, a natural tolerant variant of RERM is indeed sufficient for $γ$-tolerant robust learning VC classes over $\mathbb{R}^d$, and requires only $\tilde{O}\left( \frac{VC(H)d\log \frac{D}{γδ}}{ε^2}\right)$ samples for robustness regions of (maximum) diameter $D$.

cs.LG↗

Robust mutual synchronization in long spin Hall nano-oscillator chains

Mutual synchronization of N serially connected spintronic nano-oscillators increases their coherence by a factor $N$ and their output power by $N^2$. Increasing the number of mutually synchronized nano-oscillators in chains is hence of great importance for better signal quality and also for emerging applications such as oscillator-based neuromorphic computing and Ising machines where larger N can tackle larger problems. Here we fabricate spin Hall nano-oscillator chains of up to 50 serially connected nano-constrictions in W/NiFe, W/CoFeB/MgO, and NiFe/Pt stacks and demonstrate robust and complete mutual synchronization of up to 21 nano-constrictions, reaching linewidths of below 200 kHz and quality factors beyond 79,000, while operating at 10 GHz. We also find a square increase in the peak power with the increasing number of mutually synchronized oscillators, resulting in a factor of 400 higher peak power in long chains compared to individual nano-constrictions. Although chains longer than 21 nano-constrictions also show complete mutual synchronization, it is not as robust and their signal quality does not improve as much as they prefer to break up into partially synchronized states. The low current and low field operation of these oscillators along with their wide frequency tunability (2-28 GHz) with both current and magnetic fields, make them ideal candidates for on-chip GHz-range applications and neuromorphic computing.

cond-mat.mes-hall↗

Bounding the List Color Function Threshold from Above

The chromatic polynomial of a graph $G$, denoted $P(G,m)$, is equal to the number of proper $m$-colorings of $G$ for each $m \in \mathbb{N}$. In 1990, Kostochka and Sidorenko introduced the list color function of graph $G$, denoted $P_{\ell}(G,m)$, which is a list analogue of the chromatic polynomial. The list color function threshold of $G$, denoted $τ(G)$, is the smallest $k \geq χ(G)$ such that $P_{\ell}(G,m) = P(G,m)$ whenever $m \geq k$. It is known that for every graph $G$, $τ(G)$ is finite, and in fact, $τ(G) \leq (|E(G)|-1)/\ln(1+ \sqrt{2}) + 1$. It is also known that when $G$ is a cycle or chordal graph, $G$ is enumeratively chromatic-choosable which means $τ(G) = χ(G)$. A recent paper of Kaul et al. suggests that understanding the list color function threshold of complete bipartite graphs is essential to the study of the extremal behavior of $τ$. In this paper we show that for any $n \geq 2$, $τ(K_{2,n}) \leq \lceil (n+2.05)/1.24 \rceil$ which gives an improvement on the general upper bound for $τ(G)$ when $G = K_{2,n}$. We also develop additional tools that allow us to show that $τ(K_{2,3}) = χ(K_{2,3})$ and $τ(K_{2,4}) = τ(K_{2,5}) = 3$.

math.CO↗

Large spin Hall conductivity in epitaxial thin films of kagome antiferromagnet Mn$_3$Sn at room temperature

Mn$_3$Sn is a non-collinear antiferromagnetic quantum material that exhibits a magnetic Weyl semimetallic state and has great potential for efficient memory devices. High-quality epitaxial $c$-plane Mn$_3$Sn thin films have been grown on a sapphire substrate using a Ru seed layer. Using spin pumping induced inverse spin Hall effect measurements on $c$-plane epitaxial Mn$_3$Sn/Ni$_{80}$Fe$_{20}$, we measure spin-diffusion length ($λ_{\rm Mn_3Sn}$), and spin Hall conductivity ($σ_{\rm{SH}}$) of Mn$_3$Sn thin films: $λ_{\rm Mn_3Sn}=0.42\pm 0.04$ nm and $σ_{\rm{SH}}=-702~\hbar/ e~Ω^{-1}$cm$^{-1}$. While $λ_{\rm Mn_3Sn}$ is consistent with earlier studies, $σ_{\rm{SH}}$ is an order of magnitude higher and of the opposite sign. The behavior is explained on the basis of excess Mn, which shifts the Fermi level in our films, leading to the observed behavior. Our findings demonstrate a technique for engineering $σ_{\rm{SH}}$ of Mn$_3$Sn films by employing Mn composition for functional spintronic devices.

cond-mat.mtrl-sci↗

Large spin-to-charge conversion at the two-dimensional interface of transition metal dichalcogenides and permalloy

Spin-to-charge conversion is an essential requirement for the implementation of spintronic devices. Recently, monolayers of semiconducting transition metal dichalcogenides (TMDs) have attracted considerable interest for spin-to-charge conversion due to their high spin-orbit coupling and lack of inversion symmetry in their crystal structure. However, reports of direct measurement of spin-to-charge conversion at TMD-based interfaces are very much limited. Here, we report on the room temperature observation of a large spin-to-charge conversion arising from the interface of Ni$_{80}$Fe$_{20}$ (Py) and four distinct large area ($\sim 5\times2$~mm$^2$) monolayer (ML) TMDs namely, MoS$_2$, MoSe$_2$, WS$_2$, and WSe$_2$. We show that both spin mixing conductance and the Rashba efficiency parameter ($λ_{IREE}$) scales with the spin-orbit coupling strength of the ML TMD layers. The $λ_{IREE}$ parameter is found to range between $-0.54$ and $-0.76$ nm for the four monolayer TMDs, demonstrating a large spin-to-charge conversion. Our findings reveal that TMD/ferromagnet interface can be used for efficient generation and detection of spin current, opening new opportunities for novel spintronic devices.

physics.app-ph↗

Combining Gradients and Probabilities for Heterogeneous Approximation of Neural Networks

This work explores the search for heterogeneous approximate multiplier configurations for neural networks that produce high accuracy and low energy consumption. We discuss the validity of additive Gaussian noise added to accurate neural network computations as a surrogate model for behavioral simulation of approximate multipliers. The continuous and differentiable properties of the solution space spanned by the additive Gaussian noise model are used as a heuristic that generates meaningful estimates of layer robustness without the need for combinatorial optimization techniques. Instead, the amount of noise injected into the accurate computations is learned during network training using backpropagation. A probabilistic model of the multiplier error is presented to bridge the gap between the domains; the model estimates the standard deviation of the approximate multiplier error, connecting solutions in the additive Gaussian noise space to actual hardware instances. Our experiments show that the combination of heterogeneous approximation and neural network retraining reduces the energy consumption for multiplications by 70% to 79% for different ResNet variants on the CIFAR-10 dataset with a Top-1 accuracy loss below one percentage point. For the more complex Tiny ImageNet task, our VGG16 model achieves a 53 % reduction in energy consumption with a drop in Top-5 accuracy of 0.5 percentage points. We further demonstrate that our error model can predict the parameters of an approximate multiplier in the context of the commonly used additive Gaussian noise (AGN) model with high accuracy. Our software implementation is available under https://github.com/etrommer/agn-approx.

cs.LG↗

End-to-End Semi-Supervised Learning for Video Action Detection

In this work, we focus on semi-supervised learning for video action detection which utilizes both labeled as well as unlabeled data. We propose a simple end-to-end consistency based approach which effectively utilizes the unlabeled data. Video action detection requires both, action class prediction as well as a spatio-temporal localization of actions. Therefore, we investigate two types of constraints, classification consistency, and spatio-temporal consistency. The presence of predominant background and static regions in a video makes it challenging to utilize spatio-temporal consistency for action detection. To address this, we propose two novel regularization constraints for spatio-temporal consistency; 1) temporal coherency, and 2) gradient smoothness. Both these aspects exploit the temporal continuity of action in videos and are found to be effective for utilizing unlabeled videos for action detection. We demonstrate the effectiveness of the proposed approach on two different action detection benchmark datasets, UCF101-24 and JHMDB-21. In addition, we also show the effectiveness of the proposed approach for video object segmentation on the Youtube-VOS which demonstrates its generalization capability The proposed approach achieves competitive performance by using merely 20% of annotations on UCF101-24 when compared with recent fully supervised methods. On UCF101-24, it improves the score by +8.9% and +11% at 0.5 f-mAP and v-mAP respectively, compared to supervised approach.

cs.CV↗

RAPID: AppRoximAte Pipelined Soft Multipliers and Dividers for High-Throughput and Energy-Efficiency

The rapid updates in error-resilient applications along with their quest for high throughput have motivated designing fast approximate functional units for Field-Programmable Gate Arrays (FPGAs). Studies that proposed imprecise functional techniques are posed with three shortcomings: first, most inexact multipliers and dividers are specialized for Application-Specific Integrated Circuit (ASIC) platforms. Second, state-of-the-art (SoA) approximate units are substituted, mostly in a single kernel of a multi-kernel application. Moreover, the end-to-end assessment is adopted on the Quality of Results (QoR), but not on the overall gained performance. Finally, existing imprecise components are not designed to support a pipelined approach, which could boost the operating frequency/throughput of, e.g., division-included applications. In this paper, we propose RAPID, the first pipelined approximate multiplier and divider architecture, customized for FPGAs. The proposed units efficiently utilize 6-input Look-up Tables (6-LUTs) and fast carry chains to implement Mitchell's approximate algorithms. Our novel error-refinement scheme not only has negligible overhead over the baseline Mitchell's approach but also boosts its accuracy to 99.4% for arbitrary size of multiplication and division. Experimental results demonstrate the efficiency of the proposed pipelined and non-pipelined RAPID multipliers and dividers over accurate counterparts. Moreover, the end-to-end evaluations of RAPID, deployed in three multi-kernel applications in the domains of bio-signal processing, image processing, and moving object tracking for Unmanned Air Vehicles (UAV) indicate up to 45% improvements in area, latency, and Area-Delay-Product (ADP), respectively, over accurate kernels, with negligible loss in QoR.

cs.AR↗

Exact recovery algorithm for Planted Bipartite Graph in Semi-random Graphs

The problem of finding the largest induced balanced bipartite subgraph in a given graph is NP-hard. This problem is closely related to the problem of finding the smallest Odd Cycle Transversal. In this work, we consider the following model of instances: starting with a set of vertices $V$, a set $S \subseteq V$ of $k$ vertices is chosen and an arbitrary $d$-regular bipartite graph is added on it; edges between pairs of vertices in $S \times (V \setminus S)$ and $(V \setminus S) \times (V \setminus S)$ are added with probability $p$. Since for $d=0$, the problem reduces to recovering a planted independent set, we don't expect efficient algorithms for $k=o(\sqrt{n})$. This problem is a generalization of the planted balanced biclique problem where the bipartite graph induced on $S$ is a complete bipartite graph; [Lev18] gave an algorithm for recovering $S$ in this problem when $k=Ω(\sqrt{n})$. Our main result is an efficient algorithm that recovers (w.h.p.) the planted bipartite graph when $k=Ω_p(\sqrt{n \log n})$ for a large range of parameters. Our results also hold for a natural semi-random model of instances, which involve the presence of a monotone adversary. Our proof shows that a natural SDP relaxation for the problem is integral by constructing an appropriate solution to it's dual formulation. Our main technical contribution is a new approach for constructing the dual solution where we calibrate the eigenvectors of the adjacency matrix to be the eigenvectors of the dual matrix. We believe that this approach may have applications to other recovery problems in semi-random models as well. When $k=Ω(\sqrt{n})$, we give an algorithm for recovering $S$ whose running time is exponential in the number of small eigenvalues in graph induced on $S$; this algorithm is based on subspace enumeration techniques due to the works of [KT07,ABS10,Kol11].

cs.DS↗

Video Action Detection: Analysing Limitations and Challenges

Beyond possessing large enough size to feed data hungry machines (eg, transformers), what attributes measure the quality of a dataset? Assuming that the definitions of such attributes do exist, how do we quantify among their relative existences? Our work attempts to explore these questions for video action detection. The task aims to spatio-temporally localize an actor and assign a relevant action class. We first analyze the existing datasets on video action detection and discuss their limitations. Next, we propose a new dataset, Multi Actor Multi Action (MAMA) which overcomes these limitations and is more suitable for real world applications. In addition, we perform a biasness study which analyzes a key property differentiating videos from static images: the temporal aspect. This reveals if the actions in these datasets really need the motion information of an actor, or whether they predict the occurrence of an action even by looking at a single frame. Finally, we investigate the widely held assumptions on the importance of temporal ordering: is temporal ordering important for detecting these actions? Such extreme experiments show existence of biases which have managed to creep into existing methods inspite of careful modeling.

cs.CV↗