Search

In the cops and robber game, there are multiple cops and a single robber taking turns moving along the edges of a graph. The goal of the cops is to capture the robber (move to the same vertex as the robber) and the goal of the robber is to avoid capture. The cop number of a given graph is the smallest number of cops required to ensure the capture of the robber. The k-component order connectivity of a graph G = (V, E) is the size of a smallest set U, such that all the connected components of the induced graph on V \ U are of size at most k. In this brief note, we provide a bound on the cop number of graphs in terms of their 2-component order connectivity., Comment: 5 pages

We demonstrate that the spinning LQG black hole can act as a cosmic particle accelerator. The LQG solution is singularity-free and can possess spin greater than that of a Kerr black hole. The additional black hole hair, arising from quantum effects, significantly influences the particle dynamics around the black hole. Under suitable physical conditions, the center-of-mass energy can grow arbitrarily high during the collision of two generic particles in the spacetime of an extremal black hole. In the non-extremal case, there exists a finite upper bound on the center-of-mass energy, the maximum value of which depends on the LQG parameter. These results are particularly interesting from an astrophysical perspective, especially in the context of probing Planck-scale physics., Comment: 19 pages, 6 figures

Modern cloud has turned data services into easily accessible commodities. With just a few clicks, users are now able to access a catalog of data processing systems for a wide range of tasks. However, the cloud brings in both complexity and opportunity. While cloud users can quickly start an application by using various data services, it can be difficult to configure and optimize these services to gain the most value from them. For cloud providers, managing every aspect of an ever-increasing set of data services, while meeting customer SLAs and minimizing operational cost is becoming more challenging. Cloud technology enables the collection of significant amounts of workload traces and system telemetry. With the progress in data science (DS) and machine learning (ML), it is feasible and desirable to utilize a data-driven, ML-based approach to automate various aspects of data services, resulting in the creation of autonomous data services. This paper presents our perspectives and insights on creating autonomous data services on Azure. It also covers the future endeavors we plan to undertake and unresolved issues that still need attention., Comment: SIGMOD Companion of the 2023 International Conference on Management of Data. 2023

Automating code documentation through explanatory text can prove highly beneficial in code understanding. Large Language Models (LLMs) have made remarkable strides in Natural Language Processing, especially within software engineering tasks such as code generation and code summarization. This study specifically delves into the task of generating natural-language summaries for code snippets, using various LLMs. The findings indicate that Code LLMs outperform their generic counterparts, and zero-shot methods yield superior results when dealing with datasets with dissimilar distributions between training and testing sets., Comment: Accepted at the Forum for Information Retrieval Evaluation 2023 (IRSE Track)

This paper introduces a novel line-of-sight (LoS) $\beta-\gamma$ terahertz (THz) channel model that closely mirrors physical reality by considering radiation trapping. Our channel model provides an exhaustive modeling of the physical phenomena including the amount of re-radiation available at the receiver, parametrized by $\beta$, and the balance between scattering and noise contributions, parametrized by $\gamma$, respectively. Our findings indicate a nontrivial relationship between average limiting received signal-to-noise ratio (SNR) and distance emphasizing the significance of $\gamma$ in THz system design. We further propose new maximum likelihood (ML) thresholds for pulse amplitude modulation (PAM) and quadrature amplitude modulation (QAM) schemes, resulting in analytical symbol error rate (SER) expressions that account for different noise variances across constellation points. The results confirm that the analytical SER closely matches the true simulated SER when using an optimal detector. As expected, under maximum molecular re-radiation, the true SER is shown to be lower than that produced by a suboptimal detector that assumes equal noise variances.

The choice of the sensing matrix is crucial in compressed sensing. Random Gaussian sensing matrices satisfy the restricted isometry property, which is crucial for solving the sparse recovery problem using convex optimization techniques. However, tight-frame sensing matrices result in minimum mean-squared-error recovery given oracle knowledge of the support of the sparse vector. If the sensing matrix is not tight, could one achieve the recovery performance assured by a tight frame by suitably designing the recovery strategy? -- This is the key question addressed in this paper. We consider the analysis-sparse l1-minimization problem with a generalized l2-norm-based data-fidelity and show that it effectively corresponds to using a tight-frame sensing matrix. The new formulation offers improved performance bounds when the number of non-zeros is large. One could develop a tight-frame variant of a known sparse recovery algorithm using the proposed formalism. We solve the analysis-sparse recovery problem in an unconstrained setting using proximal methods. Within the tight-frame sensing framework, we rescale the gradients of the data-fidelity loss in the iterative updates to further improve the accuracy of analysis-sparse recovery. Experimental results show that the proposed algorithms offer superior analysis-sparse recovery performance. Proceeding further, we also develop deep-unfolded variants, with a convolutional neural network as the sparsifying operator. On the application front, we consider compressed sensing image recovery. Experimental results on Set11, BSD68, Urban100, and DIV2K datasets show that the proposed techniques outperform the state-of-the-art techniques, with performance measured in terms of peak signal-to-noise ratio and structural similarity index metric., Comment: 28 pages, 9 figures

High-dimensional data is common in multiple areas, such as health care and genomics, where the number of features can be tens of thousands. In such scenarios, the large number of features often leads to inefficient learning. Constraint generation methods have recently enabled efficient learning of L1-regularized support vector machines (SVMs). In this paper, we leverage such methods to obtain an efficient learning algorithm for the recently proposed minimax risk classifiers (MRCs). The proposed iterative algorithm also provides a sequence of worst-case error probabilities and performs feature selection. Experiments on multiple high-dimensional datasets show that the proposed algorithm is efficient in high-dimensional scenarios. In addition, the worst-case error probability provides useful information about the classifier performance, and the features selected by the algorithm are competitive with the state-of-the-art., Comment: Accepted for the 39th Conference on Uncertainty in Artificial Intelligence (UAI 2023)

Libraries for supervised classification have enabled the wide-spread usage of machine learning methods. Existing libraries, such as scikit-learn, caret, and mlpack, implement techniques based on the classical empirical risk minimization (ERM) approach. We present a Python library, MRCpy, that implements minimax risk classifiers (MRCs) based on the robust risk minimization (RRM) approach. The library offers multiple variants of MRCs that can provide performance guarantees, enable efficient learning in high dimensions, and adapt to distribution shifts. MRCpy follows an object-oriented approach and adheres to the standards of popular Python libraries, such as scikit-learn, facilitating readability and easy usage together with a seamless integration with other libraries. The source code is available under the GPL-3.0 license at https://github.com/MachineLearningBCAM/MRCpy.

This work studies the long-standing problems of model capacity and negative interference in multilingual neural machine translation MNMT. We use network pruning techniques and observe that pruning 50-70% of the parameters from a trained MNMT model results only in a 0.29-1.98 drop in the BLEU score. Suggesting that there exist large redundancies even in MNMT models. These observations motivate us to use the redundant parameters and counter the interference problem efficiently. We propose a novel adaptation strategy, where we iteratively prune and retrain the redundant parameters of an MNMT to improve bilingual representations while retaining the multilinguality. Negative interference severely affects high resource languages, and our method alleviates it without any additional adapter modules. Hence, we call it parameter-free adaptation strategy, paving way for the efficient adaptation of MNMT. We demonstrate the effectiveness of our method on a 9 language MNMT trained on TED talks, and report an average improvement of +1.36 BLEU on high resource pairs. Code will be released here.

Microsoft's internal big-data infrastructure is one of the largest in the world -- with over 300k machines running billions of tasks from over 0.6M daily jobs. Operating this infrastructure is a costly and complex endeavor, and efficiency is paramount. In fact, for over 15 years, a dedicated engineering team has tuned almost every aspect of this infrastructure, achieving state-of-the-art efficiency (>60% average CPU utilization across all clusters). Despite rich telemetry and strong expertise, faced with evolving hardware/software/workloads this manual tuning approach had reached its limit -- we had plateaued. In this paper, we present KEA, a multi-year effort to automate our tuning processes to be fully data/model-driven. KEA leverages a mix of domain knowledge and principled data science to capture the essence of our cluster dynamic behavior in a set of machine learning (ML) models based on collected system data. These models power automated optimization procedures for parameter tuning, and inform our leadership in critical decisions around engineering and capacity management (such as hardware and data center design, software investments, etc.). We combine "observational" tuning (i.e., using models to predict system behavior without direct experimentation) with judicious use of "flighting" (i.e., conservative testing in production). This allows us to support a broad range of applications that we discuss in this paper. KEA continuously tunes our cluster configurations and is on track to save Microsoft tens of millions of dollars per year. At the best of our knowledge, this paper is the first to discuss research challenges and practical learnings that emerge when tuning an exabyte-scale data infrastructure.

Employing the free energy landscape, we study the phase transition and its dynamics for a class of regular black holes in Anti-de Sitter spacetime governed by the coupling of non-linear electrodynamics, which reduces to Hayward and Bardeen solutions for particular values of spacetime parameters. The Fokker-Planck equation is solved numerically by imposing the reflecting boundary condition and a suitable initial condition, using which, we investigate the probabilistic evolution of regular AdS black holes. In this approach, the on-shell Gibbs free energy is treated as a function of the radius of the event horizon, which happens to be the order parameter of the phase transition. The numerical solution is also obtained for the absorbing boundary condition. The dynamics of switching between the coexistence small black hole phase and large black hole phase due to the thermal fluctuation is probed by calculating the first passage time. The effect of temperature on the dynamical process is also investigated., Comment: 20 pages, 10 figures. Comments are welcome

The operation of Terahertz (THz) communication can be significantly impacted by the interaction between the transmitted wave and the molecules in the atmosphere. In particular, it has been observed experimentally that the signal undergoes not only molecular absorption, but also molecular re-radiation. Two extreme modeling assumptions are prevalent in the literature, where the re-radiated energy is modeled in the first as additive Gaussian noise and in the second as a scattered component strongly correlated to the actual signal. Since the exact characterization is still an open problem, we provide in this paper the first comparative study of the performance of a reconfigurable intelligent surface (RIS) assisted THz system under these two extreme models of re-radiation. In particular, we employ an RIS to overcome the large pathloss by creating a virtual line-of-sight (LOS) path. We then develop an optimization framework for this setup and utilize the block-coordinate descent (BCD) method to iteratively optimize both RIS configuration vector and receive beamforming weight resulting in significant throughput gains for the user of interest compared to random RIS configurations. As expected, our results reveal that better throughput is achieved under the scattering assumption for the molecular re-radiation than the noise assumption.

We solve the analysis sparse coding problem considering a combination of convex and non-convex sparsity promoting penalties. The multi-penalty formulation results in an iterative algorithm involving proximal-averaging. We then unfold the iterative algorithm into a trainable network that facilitates learning the sparsity prior. We also consider quantization of the network weights. Quantization makes neural networks efficient both in terms of memory and computation during inference, and also renders them compatible for low-precision hardware deployment. Our learning algorithm is based on a variant of the ADAM optimizer in which the quantizer is part of the forward pass and the gradients of the loss function are evaluated corresponding to the quantized weights while doing a book-keeping of the high-precision weights. We demonstrate applications to compressed image recovery and magnetic resonance image reconstruction. The proposed approach offers superior reconstruction accuracy and quality than state-of-the-art unfolding techniques and the performance degradation is minimal even when the weights are subjected to extreme quantization., Comment: 8 pages + references, 7 figures and 4 tables

In this article, we study the propagation of scalar fields in the gravitational background of a spherically symmetric, electrically charged, regular-de Sitter black holes in higher dimensions. We carry out a thorough investigation of the grey body factors, by deriving analytical expressions for both minimally and non-minimally coupled scalar fields. Then we study the profile of grey body factors in terms of particle properties, namely the angular momentum quantum number and the non-minimal coupling constant, and the spacetime properties, namely the dimension, the cosmological constant and the non-linear charge parameter. We then derive the Hawking radiation spectra for a higher-dimensional electrically charged regular-de Sitter black hole. We find that the non-minimal coupling may suppress the greybody factor and the non-linear charge parameter may enhance it, but they both suppress the energy emission rate of Hawking radiation., Comment: 26 pages, 5 figures. Comments are welcome

We study the thermodynamics of Ayon Beato-Garcia black hole and the relation between the photon orbits and the thermodynamic phase transition of the black hole. Then we study the interaction between the microstructures of the black hole using Ruppeiner geometry. The radius of photon orbit and minimum impact parameter behave non-monotonously below the critical point, which mimics the behaviour of Hawking temperature and pressure in extended thermodynamics. Their change during the large black hole and small black hole phase transition serve as the order parameter, having a critical exponent $1/2$. The results show that gravity and thermodynamics of Ayon Beato-Garcia black hole are closely related. Following this we study the thermodynamic geometry which gives insight of the microstructure interaction of the black hole. We find that the large black hole phase is analogous to bosonic gas with a dominant attractive interaction, and the small black hole phase behaves like an anyonic gas with both attractive and repulsive interactions., Comment: 21 pages, 12 figures. Comments are welcome

Simplifying the detection procedure and improving the bit error rate (BER) performance of a non-coherent receiver in ambient backscatter is vital for enhancing its ability to function without the channel state information (CSI). In this work, we analyze the BER performance of Manchester encoding which is implemented at the transmitter for data transmission, and demonstrate that the optimal decision rule is independent of the system parameters. Further, through extensive numerical results, it is shown that the ambient backscatter system can achieve a signal-to-noise ratio (SNR) gain with Manchester encoding compared to the commonly used uncoded direct on-off keying (OOK) modulation, when used in conjunction with a multi-antenna receiver employing the direct-link cancellation.

Understanding the dynamic behavior of computer programs during normal working conditions is an important task, which has multiple security benefits such as the development of behavior-based anomaly detection, vulnerability discovery, and patching. Existing works achieved this goal by collecting and analyzing various data including network traffic, system calls, instruction traces, etc. In this paper, we explore the use of a new type of data, performance counters, to analyze the dynamic behavior of programs. Using existing primitives, we develop a tool named perfextract to capture data from different performance counters for a program during its startup time, thus forming multiple time series to represent the dynamic behavior of the program. We analyze the collected data and develop a semi-supervised clustering algorithm that allows us to classify each program using its performance counter time series into a specific group and to identify the intrinsic behavior of that group. We carry out extensive experiments with 18 real-world programs that belong to 4 groups including web browsers, text editors, image viewers, and audio players. The experimental results show that the examined programs can be accurately differentiated based on their performance counter data regardless of whether programs are run in physical or virtual environments., Comment: DYNAMICS 2020: DYnamic and Novel Advances in Machine Learning and Intelligent Cyber Security (DYNAMICS) Workshop

In this paper, we address the task of improving pair-wise machine translation for specific low resource Indian languages. Multilingual NMT models have demonstrated a reasonable amount of effectiveness on resource-poor languages. In this work, we show that the performance of these models can be significantly improved upon by using back-translation through a filtered back-translation process and subsequent fine-tuning on the limited pair-wise language corpora. The analysis in this paper suggests that this method can significantly improve a multilingual model's performance over its baseline, yielding state-of-the-art results for various Indian languages., Comment: ICON 2020 Short paper

We study the phase structure and the microscopic interactions in regular Bardeen AdS black hole. The stable and metastable phases in the black hole are analysed through coexistence and spinodal curves. The solutions are obtained numerically as the analytic solution to the coexistence curve is not feasible. The $P_r-T_r$ coexistence equation is obtained using a fitting formula. The coexistence and spinodal curves are plotted in $P_r-T_r$ and $T_r-V_r$ planes to explore the phase structure of the black hole. In the second part of our study, we were able to probe the microscopic interactions of regular Bardeen AdS black hole using the novel Ruppeiner geometry proposed by S.W. Wei \emph{et.al} Phys. Rev. Lett.123, 071103 (2019). It is found that the microscopic interactions are not same in the small black hole (SBH) and large black hole (LBH) phases. In the SBH phase, there exists a repulsive interaction in the microstructure in the low temperature regime. In contrast, the microstructure associated with the LBH phase has attractive interaction throughout the parameter space. We found that, along the coexistence temperature both the SBH and LBH branches diverge to negative infinity with a critical exponent equal to $1/2$., Comment: 21 pages,13 figures

Recently, a generalised solution for Einstein equations of a rotating compact body, surrounded by matter field was proposed, which is the Kerr-Newman spacetime with an anisotropic matter. The solution possesses an additional hair, along with the conventional mass, charge and spin, which arises from the negative radial pressure of the anisotropic matter. In this article we show that, this new class of black holes can act as a particle accelerator during the collision of two generic particles in its gravitational field in the ergo-region. The centre of mass energy of the particles shoots to arbitrary high value in the vicinity of event horizon for the extremal black hole. The physical conditions for the collision to take place are obtained by studying the horizon structure and circular particle motion. The results are interesting from astrophysical perspective., Comment: 22 pages, 7 figures

Modified gravity theories are of great interest in both observational and theoretical studies. In this article we study the correlation between the null geodesics in the background of a four-dimensional Einstein-Maxwell-Gauss-Bonnet AdS black hole, a modification of Einstein gravity, and it's thermodynamic phase transition. We study the phase structure of the black hole, using the coexistence and spinodal curves, to understand the phase transition in a extended phase space. The imprints of this phase transition features are observed in the behaviour of photon orbit radius and minimum impact parameter with respect to the Hawking temperature and pressure. The change in these two quantities during the phase transition serve as order parameter which characterises the critical behaviour. The correlation shows that thermodynamic phase transition can be studied by observing the effects of strong gravitational field and vice versa., Comment: 18 pages, 5 figures

Born-Infeld AdS black hole exhibits a reentrant phase transition for certain values of the Born-Infeld parameter $b$. This behaviour is an additional feature compared to the van der Waals like phase transition observed in charged AdS black holes. Therefore, it is worth observing the underlying microscopic origin of this reentrant phase transition. Depending on the value of the parameter $b$, the black hole system has four different cases: no phase transition, a reentrant phase transition with two scenarios, or a van der Waals-like (standard) phase transition. In this article, by employing a novel Ruppeiner geometry method in the parameter space of temperature and volume, we investigate the microstructure of Born-Infeld AdS black hole via the phase transition study, which includes standard and reentrant phase transition. We find that the microstructures of the black hole that lead to standard and reentrant phase transitions are distinct in nature. The standard phase transition is characterised by the typical RN-AdS microstructure. In this case, the small black hole phase has a dominant repulsive interaction for the low temperature case. Interestingly, during the reentrant phase transition, displayed by the system in a range of pressures for specific $b$ values, the dominant attractive nature of interaction in the microstructure is preserved. Our results suggest that in the reentrant phase transition case, the intermediate black holes behave like a bosonic gas, and in the standard phase transition case the small black holes behave like a quantum anyon gas. In both cases, the large black hole phase displays an interaction similar to the bosonic gas. The critical phenomenon is observed from the curvature scalar, including the signature of the reentrant phase transition., Comment: Thoroughly revised. 17 pages, 9 figures

In this article, we analytically investigate the microstructure of a five-dimensional neutral Gauss-Bonnet black hole, in the background of anti-de Sitter spacetime, using scalar curvature of the Ruppeiner geometry constructed via adiabatic compressibility. The microstructure details associated with the small-large black hole phase transition are probed in the parameter space of pressure and volume. The curvature scalar shows similar properties for both phases of the black hole, it diverges in the vicinity of critical point and approaches zero for extremal black holes. We show that the dominant interaction among black hole molecules is attractive. This study also affirms that the nature of the microstructure interaction remains unchanged during the small-large black hole phase transition, even though the microstructures are different for both phases., Comment: 14 pages, 3 figures

We demonstrate that the four-dimensional Gauss-Bonnet black hole can act as a particle accelerator with arbitrarily high centre-of-mass energy, during the collision of two general particles near the event horizon. The Gauss-Bonnet coupling constant $\alpha$, provides a deviation in the results from that of Kerr black hole. Our results show that the horizon structure, the range of allowed angular momentum and the critical angular momentum depend on the value of $\alpha$. For extremal cases, the centre-of-mass energy diverges near the horizon, suggesting that Gauss-Bonnet black hole can also act as a particle accelerator like a Kerr black hole. This is interesting in the context of probing the Planck scale physics. For the non-extremal case there exists a finite upper bound on the centre-of-mass energy, the maximal value of which depends on the parameter $\alpha$., Comment: 17 pages, 4 figures

We study the interaction between the microstructures of Hayward-AdS black hole using Ruppeiner geometry. Our investigation shows that the dominant interaction between the black hole molecules is attractive in most part of the parametric space of temperature and volume, as in van der Waals system. However, in contrast to the van der Waals fluid, there exists a weak dominant repulsive interaction for small black hole phase in some parameter range. This result clearly distinguishes the interactions in a magnetically charged black hole from that of van der Waals fluid. However, these sort of interactions are characteristic for charged black holes since they do not dependent on magnetic charge or temperature., Comment: 14 pages, 5 figures. Version accepted for publication on Physics Letters B

We explore the thermodynamic and phase transition properties of asymptotically AdS black holes within Einstein-Gauss-Bonnet gravity, focusing on Joule Thomson expansion. Thermodynamics is studied in the extended phase space, where the cosmological constant serves as thermodynamic pressure. We observe that the black hole undergoes a phase transition similar to that of a van der Waals system. We analyze charged and neutral cases separately to distinguish the effect of charge and Gauss Bonnet parameter on critical behavior and examine the phase structure. We find that the Gauss-Bonnet coupling parameter behaves similarly to black hole charge or spin, guiding the phase structure. To understand the underlying phase structure determined by the Gauss-Bonnet coefficient $\alpha$, we introduce a new order parameter. We discover that the change in the conjugate variable to the Gauss-Bonnet parameter acts as an order parameter, demonstrating a critical exponent of $1/2$ in the vicinity of the critical point. Since the phase structure is analogous to that of a van der Waals fluid, we investigate the Joule-Thomson expansion of the black hole. We analytically study the Joule-Thomson expansion, focusing on three key characteristics: the Joule-Thomson coefficient, inversion curves, and isenthalpic curves. We obtain isenthalpic curves in the $T-P$ plane and illustrate the cooling-heating regions., Comment: 23 pages, 6 figures

We study the phase transition of a regular Hayward-AdS black hole by introducing a new order parameter, the potential conjugate to the magnetic charge due to the non-linearly coupled electromagnetic field. We use Landau continuous phase transition theory to discuss the van der Waals like critical phenomena of the black hole. The popular interpretation of the AdS black hole phase transition as between a large and a small black hole is reinterpreted as the transition between a high potential phase and a low potential phase. The underlying microstructure for this phase transition is probed using the Ruppeiner geometry. By investigating the behaviour of the Ruppeiner scalar curvature, we find that the charged and uncharged (effective) molecules of the black hole have distinct microstructures analogous to fermion and boson gas., Comment: 15 pages, 2 figures

This paper focuses on the non-coherent detection in ambient backscatter communication, which is highly appealing for systems where the trade-off between signaling overhead and the actual data transmission is very critical. Modeling the time-selective fading channel as a first-order autoregressive (AR) process, we propose a new receiver architecture based on the direct averaging of the received signal samples for detection, which departs significantly from the energy averaging-based receivers considered in the literature. For the proposed setup, we characterize the exact asymptotic bit error rate (BER) for both single-antenna (SA) and multi-antenna (MA) receivers, and demonstrate the robustness of the new architecture to timing errors. Our results demonstrate that while the direct-link (DL) interference from the ambient power source leads to a BER floor in the SA receiver, the MA receiver can remove this interference by estimating the angle of arrival (AoA) of the DL. The analysis further quantifies the effect of improved angular resolution on the BER as a function of the number of receive antennas. A key intermediate result of our analysis is the derivation of a new concentration result for a general sum sequence that is central to the derivation of the conditional distributions of the received signal.

The success of Internet-of-Things (IoT) paradigm relies on, among other things, developing energy-efficient communication techniques that can enable information exchange among billions of battery-operated IoT devices. With its technological capability of simultaneous information and energy transfer, ambient backscatter is quickly emerging as an appealing solution for this communication paradigm, especially for the links with low data rate requirement. In this paper, we study signal detection and characterize exact bit error rate for the ambient backscatter system. In particular, we formulate a binary hypothesis testing problem at the receiver and analyze system performance under three detection techniques: a) mean threshold (MT), b) maximum likelihood threshold (MLT), and c) approximate MLT. Motivated by the energy-constrained nature of IoT devices, we perform the above analyses for two receiver types: i) the ones that can accurately track channel state information (CSI), and ii) the ones that cannot. Two main features of the analysis that distinguish this work from the prior art are the characterization of the exact conditional density functions of the average received signal energy, and the characterization of exact average bit error rate (BER) for this setup. The key challenge lies in the handling of correlation between channel gains of two hypotheses for the derivation of joint probability distribution of magnitude squared channel gains that is needed for the BER analysis.

Hardware accelerators are being increasingly deployed to boost the performance and energy efficiency of deep neural network (DNN) inference. In this paper we propose Thundervolt, a new framework that enables aggressive voltage underscaling of high-performance DNN accelerators without compromising classification accuracy even in the presence of high timing error rates. Using post-synthesis timing simulations of a DNN accelerator modeled on the Google TPU, we show that Thundervolt enables between 34%-57% energy savings on state-of-the-art speech and image recognition benchmarks with less than 1% loss in classification accuracy and no performance loss. Further, we show that Thundervolt is synergistic with and can further increase the energy efficiency of commonly used run-time DNN pruning techniques like Zero-Skip.