Designing Energy-Efficient Convolutional Neural Networks using Energy-Aware Pruning
Yang, Tien-Ju; Chen, Yu-Hsin; Sze, Vivienne
Deep convolutional neural networks (CNNs) are indispensable to state-of-the-art computer vision algorithms. However, they are still rarely deployed on battery-powered mobile devices, such as smartphones and wearable gadgets, where vision algorithms can enable many revolutionary real-world applications. The key limiting factor is the high energy consumption of CNN processing due to its high computational complexity. While there are many previous efforts that try to reduce the CNN model size or the amount of computation, we find that they do not necessarily result in lower energy consumption. Therefore, these targets do not serve as a good metric for energy cost estimation. To close the gap between CNN design and energy consumption optimization, we propose an energy-aware pruning algorithm for CNNs that directly uses the energy consumption of a CNN to guide the pruning process. The energy estimation methodology uses parameters extrapolated from actual hardware measurements. The proposed layer- by-layer pruning algorithm also prunes more aggressively than previously proposed pruning methods by minimizing the error in the output feature maps instead of the filter weights. For each layer, the weights are first pruned and then locally fine-tuned with aclosed-form least-square solution to quickly restore the accuracy. After all layers are pruned, the entire network is globally fine-tuned using back-propagation. With the proposed pruning method, the energy consumption of AlexNet and GoogLeNet is reduced by 3.7X and 1.6X, respectively, with less than 1% top-5 accuracy loss. We also show that reducing the number of target classes in AlexNet greatly decreases the number of weights, but has a limited impact on energy consumption.
Designing Energy-Efficient Convolutional Neural Networks using Energy-Aware Pruning
↧
↧
The crucial effect of early-stage gelation on the mechanical properties of cement hydrates
The crucial effect of early-stage gelation on the mechanical properties of cement hydrates
Kanduč, Matej; Li, Lunna; Frenkel, Daan; Dobnikar, Jure; Del Gado, Emanuela; Ioannidou, Aikaterini
Gelation and densification of calcium–silicate–hydrate take place during cement hydration. Both processes are crucial for the development of cement strength, and for the long-term evolution of concrete structures. However, the physicochemical environment evolves during cement formation, making it difficult to disentangle what factors are crucial for the mechanical properties. Here we use Monte Carlo and Molecular Dynamics simulations to study a coarse-grained model of cement formation, and investigate the equilibrium and arrested states. We can correlate the various structures with the time evolution of the interactions between the nano-hydrates during the preparation of cement. The novel emerging picture is that the changes of the physicochemical environment, which dictate the evolution of the effective interactions, specifically favour the early gel formation and its continuous densification. Our observations help us understand how cement attains its unique strength and may help in the rational design of the properties of cement and related materials.
↧
Polaron effects on the performance of light-harvesting systems: a quantum heat engine perspective
Polaron effects on the performance of light-harvesting systems: a quantum heat engine perspective
Zhao, Yang; Cao, Jianshu; Wang, Chen; Xu, Dazhi
We explore energy transfer in a generic three-level system, which is coupled to three non-equilibrium baths. Built on the concept of quantum heat engine, our three-level model describes non-equilibrium quantum processes including light-harvesting energy transfer, nano-scale heat transfer, photo-induced isomerization, and photovoltaics in double quantum-dots. In the context of light-harvesting, the excitation energy is first pumped up by sunlight, then is transferred via two excited states which are coupled to a phonon bath, and finally decays to the reaction center. The efficiency of this process is evaluated by steady state analysis via a polaron-transformed master equation; thus the entire range of the system-phonon coupling strength can be covered. We show that the coupling with the phonon bath not only modifies the steady state, resulting in population inversion, but also introduces a finite steady state coherence which optimizes the energy transfer flux and efficiency. In the strong coupling limit, the steady state coherence disappears and the efficiency recovers the heat engine limit given by Scovil and Schultz-Dubois (1959 Phys. Rev. Lett. 2 262).
↧
Hybrid Microscopy: Enabling Inexpensive High-Performance Imaging through Combined Physical and Optical Magnifications
Hybrid Microscopy: Enabling Inexpensive High-Performance Imaging through Combined Physical and Optical Magnifications
Zhang, Yu Shrike; Chang, Jae-Byum; Alvarez, Mario Moises; Trujillo de Santiago, Grissel; Aleman, Julio; Batzaya, Byambaa; Krishnadoss, Vaishali; Ramanujam, Aishwarya Aravamudhan; Kazemzadeh-Narbat, Mehdi; Chen, Fei; Tillberg, Paul W.; Dokmeci, Mehmet R.; Boyden, Edward; Khademhosseini, Alireza
To date, much effort has been expended on making high-performance microscopes through better instrumentation. Recently, it was discovered that physical magnification of specimens was possible, through a technique called expansion microscopy (ExM), raising the question of whether physical magnification, coupled to inexpensive optics, could together match the performance of high-end optical equipment, at a tiny fraction of the price. Here we show that such “hybrid microscopy” methods—combining physical and optical magnifications—can indeed achieve high performance at low cost. By physically magnifying objects, then imaging them on cheap miniature fluorescence microscopes (“mini-microscopes”), it is possible to image at a resolution comparable to that previously attainable only with benchtop microscopes that present costs orders of magnitude higher. We believe that this unprecedented hybrid technology that combines expansion microscopy, based on physical magnification, and mini-microscopy, relying on conventional optics—a process we refer to as Expansion Mini-Microscopy (ExMM)—is a highly promising alternative method for performing cost-effective, high-resolution imaging of biological samples. With further advancement of the technology, we believe that ExMM will find widespread applications for high-resolution imaging particularly in research and healthcare scenarios in undeveloped countries or remote places.
↧
Talking about Data: Sharing Richly Structured Information through Blogs and Wikis
Talking about Data: Sharing Richly Structured Information through Blogs and Wikis
Benson, Edward Oscar; Marcus, Adam; Howahl, Fabian; Karger, David R
Several projects have brought rich data semantics to collaborative wikis, but blogging platforms remain primarily limited to text. As blogs comprise a significant portion of the web’s content, engagement of the blogging community is crucial to the development of the semantic web. We provide a study of blog content to show a latent need for better data publishing and visualization support in blogging software. We then present DataPress, an extension to the WordPress blogging platform that enables users to publish, share, aggregate, and visualize structured information using the same workflow that they already apply to text-based content. In particular, we aim to preserve those attributes that make blogs such a successful publication medium: one-click access to the information, one-click publishing of it, natural authoring interfaces, and easy copy and paste of information (and visualizations) from other sources. We reflect on how our designs make progress toward these goals with a study of how users who installed DataPress made use of various features.
↧
↧
TURBULENT CHEMICAL DIFFUSION IN CONVECTIVELY BOUNDED CARBON FLAMES
TURBULENT CHEMICAL DIFFUSION IN CONVECTIVELY BOUNDED CARBON FLAMES
Lecoanet, Daniel; Schwab, Josiah; Quataert, Eliot; Bildsten, Lars; Timmes, F. X.; Vasil, Geoffrey M.; Oishi, Jeffrey S.; Brown, Benjamin P.; Burns, Keaton James
It has been proposed that mixing induced by convective overshoot can disrupt the inward propagation of carbon deflagrations in super-asymptotic giant branch stars. To test this theory, we study an idealized model of convectively bounded carbon flames with 3D hydrodynamic simulations of the Boussinesq equations using the pseudo-spectral code Dedalus. Because the flame propagation timescale is much longer than the convection timescale, we approximate the flame as fixed in space, and only consider its effects on the buoyancy of the fluid. By evolving a passive scalar field, we derive a turbulent chemical diffusivity produced by the convection as a function of height, D[subscript t][subscript z]). Convection can stall a flame if the chemical mixing timescale, set by the turbulent chemical diffusivity, D[subscript t], is shorter than the flame propagation timescale, set by the thermal diffusivity, κ, i.e., when D[subscript t]
↧
Controlled fabrication of nanoscale gaps using stiction
Controlled fabrication of nanoscale gaps using stiction
Niroui, Farnaz; Sletten, Ellen M.; Deotare, Parag Bhaskar; Wang, Annie I; Swager, Timothy M; Lang, Jeffrey H; Bulovic, Vladimir
Utilizing stiction, a common failure mode in micro/nano electromechanical systems (M/NEMS), we propose a method for the controlled fabrication of nanometer-thin gaps between electrodes. In this approach, a single lithography step is used to pattern cantilevers that undergo lateral motion towards opposing stationary electrodes separated by a defined gap. Upon wet developing of the pattern, capillary forces induce cantilever deflection and collapse leading to permanent adhesion between the tip and an opposing support structure. The deflection consequently reduces the separation gap between the cantilever and the electrodes neighboring the point of stiction to dimensions smaller than originally patterned. Through nanoscale force control achieved by altering device design, we demonstrate the fabrication of nanogaps having controlled widths smaller than 15 nm. We further discuss optimization of these nanoscale gaps for applications in NEM and molecular devices.
↧
Efficient distributed sensing using adaptive censoring-based inference
Efficient distributed sensing using adaptive censoring-based inference
Mu, Beipeng; Chowdhary, Girish; How, Jonathan P
In many distributed sensing applications with limited resources, it is likely that only a few agents will have valuable information at any given time. Therefore it is important to ensure that the resources are spent on communicating valuable information from informative agents. This paper presents communication-efficient distributed sensing algorithms that avoid network cluttering by having only agents with high Value of Information (VoI) broadcast their measurements to the network, while others censor themselves. A novel contribution of the presented distributed estimation algorithm is the use of an adaptively adjusted VoI threshold to determine which agents are informative. This adaptation enables the team to better balance between the communication cost incurred and the long-term accuracy of the estimation. Theoretical results are presented establishing the almost sure convergence of the communication cost and estimation error for distributions in the exponential family. Furthermore, validation through numerical simulations and real datasets shows that the new VoI-based algorithms can yield improved parameter estimates than those achieved by previously published hyperparameter consensus algorithms while incurring only a fraction of the communication cost.
↧
Stereospecific Ring-Opening Metathesis Polymerization (ROMP) of Norbornene and Tetracyclododecene by Mo and W Initiators
Stereospecific Ring-Opening Metathesis Polymerization (ROMP) of Norbornene and Tetracyclododecene by Mo and W Initiators
Autenrieth, Benjamin; Schrock, Richard Royce
We report the synthesis of >98% cis,isotactic and cis,syndiotactic polynorbornene (poly(NBE)) and poly(endo,anti-tetracyclododecene) (poly(TCD)). Cis,isotactic poly(NBE) and poly(TCD) were prepared employing Mo-based biphenolate imido alkylidene initiators, Mo(NR)(CHCMe[subscript 2]Ph)(Biphen) (Biphen = e.g., 3,3′-(t-Bu)[subscript 2]-5,5′-6,6′-(CH[subscript 3])[subscript 4]-1,1′-biphenyl-2,2′-diolate), while cis,syndiotactic poly(NBE) and poly(TCD) were prepared employing W-based imido or oxo monoaryloxide pyrrolide (MAP) initiators, W(X)(CHR′)(Pyrrolide)(OTer) (X = NR or O; OTer = a 2,6-terphenoxide). Addition of 1-hexene or coordinating solvents such as THF do not decrease the stereospecificity of the polymerization. Cis,iso and cis,syndio dyads can be distinguished through examination of [superscript 1]H and [superscript 13]C NMR spectra of the two polymers in a mixture. The polymers were hydrogenated to give isotactic and syndiotactic H-poly(NBE) and H-poly(TCD).
↧
↧
Characterization of In-Pipe Acoustic Wave for Water Leak Detection
Characterization of In-Pipe Acoustic Wave for Water Leak Detection
Khalifa, Atia E.; Ben-Mansour, Rached; Youcef-Toumi, Kamal; Choi, Changrak
This paper presents experimental observations on the characteristics of the acoustic signal propagation and attenuation inside water-filled pipes. An acoustic source (exciter) is mounted on the internal pipe wall, at a fixed location, and produces a tonal sound to simulate a leak noise with controlled frequency and amplitude, under different flow conditions. A hydrophone is aligned with the pipe centerline and can be re-positioned to capture the acoustic signal at different locations. Results showed that the wave attenuation depends on the source frequency and the line pressure. High frequency signals get attenuated more with increasing distance from the source. The optimum location to place the hydrophone for capturing the acoustic signal is not at the vicinity of source location. The optimum location also depends on the frequency and line pressure. It was also observed that the attenuation of the acoustic waves is higher in more flexible pipes like PVC ones.
↧
Pitch-Responsive Cortical Regions in Congenital Amusia
Pitch-Responsive Cortical Regions in Congenital Amusia
Albouy, Philippe; Tillmann, Barbara; Caclin, Anne; Norman-Haignere, Samuel Victor; McDermott, Joshua H.; Kanwisher, Nancy
Congenital amusia is a lifelong deficit in music perception thought to reflect an underlying impairment in the perception and memory of pitch. The neural basis of amusic impairments is actively debated. Some prior studies have suggested that amusia stems from impaired connectivity between auditory and frontal cortex. However, it remains possible that impairments in pitch coding within auditory cortex also contribute to the disorder, in part because prior studies have not measured responses from the cortical regions most implicated in pitch perception in normal individuals. We addressed this question by measuring fMRI responses in 11 subjects with amusia and 11 age- and education-matched controls to a stimulus contrast that reliably identifies pitch-responsive regions in normal individuals: harmonic tones versus frequency-matched noise. Our findings demonstrate that amusic individuals with a substantial pitch perception deficit exhibit clusters of pitch-responsive voxels that are comparable in extent, selectivity, and anatomical location to those of control participants. We discuss possible explanations for why amusics might be impaired at perceiving pitch relations despite exhibiting normal fMRI responses to pitch in their auditory cortex: (1) individual neurons within the pitch-responsive region might exhibit abnormal tuning or temporal coding not detectable with fMRI, (2) anatomical tracts that link pitch-responsive regions to other brain areas (e.g., frontal cortex) might be altered, and (3) cortical regions outside of pitch-responsive cortex might be abnormal. The ability to identify pitch-responsive regions in individual amusic subjects will make it possible to ask more precise questions about their role in amusia in future work.
SIGNIFICANCE STATEMENT The neural causes of congenital amusia, a lifelong deficit in pitch and music perception, are not fully understood. We tested the hypothesis that amusia is due to abnormalities in brain regions that respond selectively to sounds with a pitch in normal listeners. Surprisingly, amusic individuals exhibited pitch-responsive regions that were similar to normal-hearing controls in extent, selectivity, and anatomical location. We discuss how our results inform current debates on the neural basis of amusia and how the ability to identify pitch-responsive regions in amusic subjects will make it possible to ask more precise questions about their role in amusic deficits.
↧
Resistive Switching Mechanisms on TaO[subscript x] and SrRuO[subscript 3] Thin-Film Surfaces Probed by Scanning Tunneling Microscopy
Resistive Switching Mechanisms on TaO[subscript x] and SrRuO[subscript 3] Thin-Film Surfaces Probed by Scanning Tunneling Microscopy
Moors, Marco; Wedig, Anja; Bäumer, Christoph; Skaja, Katharina; Arndt, Benedikt; Dittmann, Regina; Waser, Rainer; Valov, Ilia; Adepalli, Kiran Kumar; Lu, Qiyang; Tuller, Harry L; Yildiz, Bilge
The local electronic properties of tantalum oxide (TaO[subscript x], 2 ≤ x ≤ 2.5) and strontium ruthenate (SrRuO[subscript 3]) thin-film surfaces were studied under the influence of electric fields induced by a scanning tunneling microscope (STM) tip. The switching between different redox states in both oxides is achieved without the need for physical electrical contact by controlling the magnitude and polarity of the applied voltage between the STM tip and the sample surface. We demonstrate for TaO[subscript x] films that two switching mechanisms operate. Reduced tantalum oxide shows resistive switching due to the formation of metallic Ta, but partial oxidation of the samples changes the switching mechanism to one mediated mainly by oxygen vacancies. For SrRuO[subscript 3], we found that the switching mechanism depends on the polarity of the applied voltage and involves formation, annihilation, and migration of oxygen vacancies. Although TaO[subscript x] and SrRuO[subscript 3] differ significantly in their electronic and structural properties, the resistive switching mechanisms could be elaborated based on STM measurements, proving the general capability of this method for studying resistive switching phenomena in different classes of transition metal oxides.
↧
Theoretical Analysis of Track Generation in 3d Method of Characteristics
Theoretical Analysis of Track Generation in 3d Method of Characteristics
Shaner, Samuel Christopher; Gunow, Geoffrey Alexander; Forget, Benoit Robert Yves; Smith, Kord S.
Generating the tracks to use in a 3D Method of Characteristics (MOC) simulation is not a trivial task. The method used to generate tracks has significant implications on the memory and compute requirements for a problem and the current track generation methods have shortcomings. In this study, we provide a detailed description and analysis of the current state of the art method to generate tracks for direct 3D MOC, the Modular Ray Tracing (MRT) method. Additionally, a new global method for
generating tracks is presented that is generalizable to many geometries, domain decomposition schemes, and quadrature sets. The main difference between the global and modular track generation approaches is that the global approach does not require any knowledge of the underlying geometry discretization and is therefore more flexible in domain decomposing the geometry. Some considerations with memory
requirements and general applicability that we and others have found are discussed
↧
↧
Are You Your Friends’ Friend? Poor Perception of Friendship Ties Limits the Ability to Promote Behavioral Change
Are You Your Friends’ Friend? Poor Perception of Friendship Ties Limits the Ability to Promote Behavioral Change
Almaatouq, Abdullah; Radaelli, Laura; Pentland, Alex Paul; Shmueli, Erez
Persuasion is at the core of norm creation, emergence of collective action, and solutions to ‘tragedy of the commons’ problems. In this paper, we show that the directionality of friendship ties affect the extent to which individuals can influence the behavior of each other. Moreover, we find that people are typically poor at perceiving the directionality of their friendship ties and that this can significantly limit their ability to engage in cooperative arrangements. This could lead to failures in establishing compatible norms, acting together, finding compromise solutions, and persuading others to act. We then suggest strategies to overcome this limitation by using two topological characteristics of the perceived friendship network. The findings of this paper have significant consequences for designing interventions that seek to harness social influence for collective action.
↧
Neural Tuning Size in a Model of Primate Visual Processing Accounts for Three Key Markers of Holistic Face Processing
Neural Tuning Size in a Model of Primate Visual Processing Accounts for Three Key Markers of Holistic Face Processing
Tan, Cheston; Poggio, Tomaso A
Faces are an important and unique class of visual stimuli, and have been of interest to neuroscientists for many years. Faces are known to elicit certain characteristic behavioral markers, collectively labeled “holistic processing”, while non-face objects are not processed holistically. However, little is known about the underlying neural mechanisms. The main aim of this computational simulation work is to investigate the neural mechanisms that make face processing holistic. Using a model of primate visual processing, we show that a single key factor, “neural tuning size”, is able to account for three important markers of holistic face processing: the Composite Face Effect (CFE), Face Inversion Effect (FIE) and Whole-Part Effect (WPE). Our proof-of-principle specifies the precise neurophysiological property that corresponds to the poorly-understood notion of holism, and shows that this one neural property controls three classic behavioral markers of holism. Our work is consistent with neurophysiological evidence, and makes further testable predictions. Overall, we provide a parsimonious account of holistic face processing, connecting computation, behavior and neurophysiology.
↧
Edges and linearization: A reply
Edges and linearization: A reply
Trinh, Tue Huu
This is my reply to the commentaries on Trinh (2009). I thank the commentators – Enoch Aboh, Josef Bayer, Nigel Duffield, Roland Hinterhölzl, Anders Holmberg, Shinichiro Ishihara and Gereon Müller – for their helpful critiques, which point out several shortcomings of the target paper, opened up areas where new predictions can be made and raised questions to be investigated in future research. It is unfortunate that these critiques and this reply did not appear in the same volume. The first section of the reply will hence be a brief summary of my proposal. The theory will be presented in a slightly different way, more transparent than the original formulation but equivalent with it in content. It is my hope that the reformulation will address some of the questions raised in the commentaries. Other issues are discussed in the sections that follow
↧
Complementing quantitative typology with behavioral approaches: Evidence for typological universals
Complementing quantitative typology with behavioral approaches: Evidence for typological universals
Jaeger, T. Florian; Tily, Harry J.
Two main classes of theory have been advanced to explain correlations between linguistic features like those observed by Greenberg (1963). arbitrary constraint theories argue that certain sets of features patterm together because they have a single underlying cause in the innate language faculty (e.g., the Principles and Parameters program; see Chomsky & Lasnik 1993). functional theories argue that languages are less likely to have certain combinations of properties because, although possible in principle, they are harder to learn or to process, or less suitable for efficient communication (Hockett 1960, Bates & MacWhinney 1989, Hawkins 2004, Dryer 2007, Christiansen & Chater 2008; for further discussion see Hawkins 2007 and Jaeger & Tily 2011). The failure of Dunn, Greenhill, Levinson & Gray (2011) to find systematic feature correlations using their novel computational phylogenetic methods calls into question both of these classes of theory.
↧
↧
The Generalized Empirical Interpolation Method: Stability theory on Hilbert spaces with an application to the Stokes equation
The Generalized Empirical Interpolation Method: Stability theory on Hilbert spaces with an application to the Stokes equation
Maday, Y.; Mula, O.; Patera, Anthony T; Yano, Masayuki
The Generalized Empirical Interpolation Method (GEIM) is an extension first presented by Maday and Mula in Maday and Mula (2013) in 2013 of the classical empirical interpolation method (presented in 2004 by Barrault, Maday, Nguyen and Patera in Barrault et al. (2004)) where the evaluation at interpolating points is replaced by the more practical evaluation at interpolating continuous linear functionals on a class of Banach spaces. As outlined in Maday and Mula (2013), this allows to relax the continuity constraint in the target functions and expand both the application domain and the stability of the approach. In this paper, we present a thorough analysis of the concept of stability condition of the generalized interpolant (the Lebesgue constant) by relating it to an inf–sup problem in the case of Hilbert spaces. In the second part of the paper, it will be explained how GEIM can be employed to monitor in real time physical experiments by providing an online accurate approximation of the phenomenon that is computed by combining the acquisition of a minimal number, optimally placed, measurements from the processes with their mathematical models (parameter-dependent PDEs). This idea is illustrated through a parameter dependent Stokes problem in which it is shown that the pressure and velocity fields can efficiently be reconstructed with a relatively low-dimensional interpolation space.
↧
Differential branching fraction and angular analysis of the B + → K + μ + μ − decay
Differential branching fraction and angular analysis of the B + → K + μ + μ − decay
The LHCb Collaboration; Aaij, A.; Beteta, Abellan; Adeva, B.; Williams, Michael
The angular distribution and differential branching fraction of the decay B+→ K+μ+μ− are studied with a dataset corresponding to 1.0fb−1 of integrated luminosity, collected by the LHCb experiment. The angular distribution is measured in bins of dimuon invariant mass squared and found to be consistent with Standard Model expectations. Integrating the differential branching fraction over the full dimuon invariant mass range yields a total branching fraction of BB (B+ → K+μ+μ−) = (4.36 ± 0.15 ± 0.18) × 10−7. These measurements are the most precise to date of the B+ → K+μ+μ− decay.
↧
Pupil slicer design for the NASA-NSF extreme precision Doppler spectrograph concept WISDOM
Pupil slicer design for the NASA-NSF extreme precision Doppler spectrograph concept WISDOM
Simcoe, Robert A; Egan, Mark; Furesz, Gabor
The WIYN Spectrograph for Doppler Monitoring (WISDOM) was a concept responding to NASA's solicitation for an extreme precision radial velocity instrument for the 3.5 meter WIYN telescope on Kitt Peak in Arizona. In order to meet the spectral resolution requirement of R = 110,000 while maintaining good throughput and a manageable beam diameter, the front end design of the instrument employed a pupil slicing technique wherein a collimated beam is sliced and fed to six separate fibers. This paper presents the optical and mechanical design of the pupil slicer subassembly, a unique method of dealing with thermally induced defocus error, and the methods and results of aligning a prototype.
↧
More Pages to Explore .....
