Search

Processing-in-memory (PIM) architecture is an inherent match for data analytics application, but we observe major challenges to address when accelerating it using PIM. In this paper, we propose Darwin, a practical LRDIMM-based multi-level PIM architecture for data analytics, which fully exploits the internal bandwidth of DRAM using the bank-, bank group-, chip-, and rank-level parallelisms. Considering the properties of data analytics operators and DRAM's area constraints, Darwin maximizes the internal data bandwidth by placing the PIM processing units, buffers, and control circuits across the hierarchy of DRAM. More specifically, it introduces the bank processing unit for each bank in which a single instruction multiple data (SIMD) unit handles regular data analytics operators and bank group processing unit for each bank group to handle workload imbalance in the condition-oriented data analytics operators. Furthermore, Darwin supports a novel PIM instruction architecture that concatenates instructions for multiple thread executions on bank group processing entities, addressing the command bottleneck by enabling separate control of up to 512 different in-memory processing units simultaneously. We build a cycle-accurate simulation framework to evaluate Darwin with various DRAM configurations, optimization schemes and workloads. Darwin achieves up to 14.7x speedup over the non-optimized version. Finally, the proposed Darwin architecture achieves 4.0x-43.9x higher throughput and reduces energy consumption by 85.7% than the baseline CPU system (Intel Xeon Gold 6226 + 4 channels of DDR4-2933). Compared to the state-of-the-art PIM, Darwin achieves up to 7.5x and 7.1x in the basic query operators and TPC-H queries, respectively. Darwin is based on the latest GDDR6 and requires only 5.6% area overhead, suggesting a promising PIM solution for the future main memory system., Comment: 14 pages, 16 figures

The Event Horizon Telescope (EHT) observed in 2017 the supermassive black hole at the center of the Milky Way, Sagittarius A* (Sgr A*), at a frequency of 228.1 GHz ($\lambda$=1.3 mm). The fundamental physics tests that even a single pulsar orbiting Sgr A* would enable motivate searching for pulsars in EHT datasets. The high observing frequency means that pulsars - which typically exhibit steep emission spectra - are expected to be very faint. However, it also negates pulse scattering, an effect that could hinder pulsar detections in the Galactic Center. Additionally, magnetars or a secondary inverse Compton emission could be stronger at millimeter wavelengths than at lower frequencies. We present a search for pulsars close to Sgr A* using the data from the three most-sensitive stations in the EHT 2017 campaign: the Atacama Large Millimeter/submillimeter Array, the Large Millimeter Telescope and the IRAM 30 m Telescope. We apply three detection methods based on Fourier-domain analysis, the Fast-Folding-Algorithm and single pulse search targeting both pulsars and burst-like transient emission; using the simultaneity of the observations to confirm potential candidates. No new pulsars or significant bursts were found. Being the first pulsar search ever carried out at such high radio frequencies, we detail our analysis methods and give a detailed estimation of the sensitivity of the search. We conclude that the EHT 2017 observations are only sensitive to a small fraction ($\lesssim$2.2%) of the pulsars that may exist close to Sgr A*, motivating further searches for fainter pulsars in the region., Comment: 33 pages, 7 figures, 6 Tables. Accepted for publication in ApJ

Computerized Adaptive Testing (CAT) is a widely used, efficient test mode that adapts to the examinee's proficiency level in the test domain. CAT requires pre-trained item profiles, for CAT iteratively assesses the student real-time based on the registered items' profiles, and selects the next item to administer using candidate items' profiles. However, obtaining such item profiles is a costly process that involves gathering a large, dense item-response data, then training a diagnostic model on the collected data. In this paper, we explore the possibility of leveraging response data collected in the CAT service. We first show that this poses a unique challenge due to the inherent selection bias introduced by CAT, i.e., more proficient students will receive harder questions. Indeed, when naively training the diagnostic model using CAT response data, we observe that item profiles deviate significantly from the ground-truth. To tackle the selection bias issue, we propose the user-wise aggregate influence function method. Our intuition is to filter out users whose response data is heavily biased in an aggregate manner, as judged by how much perturbation the added data will introduce during parameter estimation. This way, we may enhance the performance of CAT while introducing minimal bias to the item profiles. We provide extensive experiments to demonstrate the superiority of our proposed method based on the three public datasets and one dataset that contains real-world CAT response data., Comment: CIKM 2023

Since wearable linkage mechanisms could control the moment transmission from actuator(s) to wearers, they can help ensure that even low-cost wearable systems provide advanced functionality tailored to users' needs. For example, if a hip mechanism transforms an input torque into a spatially-varying moment, a wearer can get effective assistance both in the sagittal and frontal planes during walking, even with an affordable single-actuator system. However, due to the combinatorial nature of the linkage mechanism design space, the topologies of such nonlinear-moment-generating mechanisms are challenging to determine, even with significant computational resources and numerical data. Furthermore, on-premise production development and interactive design are nearly impossible in conventional synthesis approaches. Here, we propose an innovative autonomous computational approach for synthesizing such wearable robot mechanisms, eliminating the need for exhaustive searches or numerous data sets. Our method transforms the synthesis problem into a gradient-based optimization problem with sophisticated objective and constraint functions while ensuring the desired degree of freedom, range of motion, and force transmission characteristics. To generate arbitrary mechanism topologies and dimensions, we employed a unified ground model. By applying the proposed method for the design of hip joint mechanisms, the topologies and dimensions of non-series-type hip joint mechanisms were obtained. Biomechanical simulations validated its multi-moment assistance capability, and its wearability was verified via prototype fabrication. The proposed design strategy can open a new way to design various wearable robot mechanisms, such as shoulders, knees, and ankles., Comment: 28 pages, 7 figures, Supplementary Materials

The phase distribution in a Bose-Einstein condensate can realize various topological states classified by distinct winding numbers. While states with different winding numbers are topologically protected in the linear Schr\"odinger equation, when nonlinearities are introduced, violations of the topological protection can occur, leading to unwinding. Exciton-polariton condensates constitute a nonlinear open-dissipative system that is well suited to studying such physics. Here we show that a one-dimensional array of exciton-polariton condensates displays a spontaneous phase unwinding from a $\pi$- to zero-state. We clarify that this collective mode transition is caused by the combined effect of nonlinearity and topological defects in the condensates. While the mode-switching phenomenon observed in our previous experiment was interpreted as the single-particle mode competition, we offer an alternative explanation in terms the collective phase unwinding and find its evidence by reanalyzing the experimental data. Our results open a route towards active control of the mode switching by manipulating the topological defects in prospective quantum polaritonic devices., Comment: 11 pages, 5 figures

In the last decade, motivated by the concept of Planckian relaxation and the possible existence of a quantum critical point in cuprate materials, holographic techniques have been extensively used to tackle the problem of strange metals and high-$T_c$ superconductors. Among the various setups, the linear axion Gubser-Rocha model has often been considered as a promising holographic model for strange metals since endowed with the famous linear in $T$ resistivity property. As fiercely advocated by Phil Anderson, beyond $T$-linear resistivity, there are several additional anomalies unique to the strange metal phase, as for example a Fermi liquid like Hall angle -- the famous problem of the two relaxation scales. In this short note, we show that the linear axion holographic Gubser-Rocha model, which presents a single momentum relaxation time, fails in this respect and therefore is not able to capture the transport phenomenology of strange metals. We prove our statement by means of a direct numerical computation, a previously demonstrated scaling analysis and also a hydrodynamic argument. Finally, we conclude with an optimistic discussion on the possible improvements and generalizations which could lead to a holographic model for strange metals in all their glory., Comment: v2: minor revisions, matching the published version

Traffic forecasting is a highly challenging task owing to the dynamical spatio-temporal dependencies of traffic flows. To handle this, we focus on modeling the spatio-temporal dynamics and propose a network termed Edge Squeeze Graph Convolutional Network (ESGCN) to forecast traffic flow in multiple regions. ESGCN consists of two modules: W-module and ES module. W-module is a fully node-wise convolutional network. It encodes the time-series of each traffic region separately and decomposes the time-series at various scales to capture fine and coarse features. The ES module models the spatio-temporal dynamics using Graph Convolutional Network (GCN) and generates an Adaptive Adjacency Matrix (AAM) with temporal features. To improve the accuracy of AAM, we introduce three key concepts. 1) Using edge features to directly capture the spatiotemporal flow representation among regions. 2) Applying an edge attention mechanism to GCN to extract the AAM from the edge features. Here, the attention mechanism can effectively determine important spatio-temporal adjacency relations. 3) Proposing a novel node contrastive loss to suppress obstructed connections and emphasize related connections. Experimental results show that ESGCN achieves state-of-the-art performance by a large margin on four real-world datasets (PEMS03, 04, 07, and 08) with a low computational cost., Comment: 7 Pages, 3 figures

We investigate the mechanics of the proton by examining the flavor-decomposed proton cosmological constants and generalized vector form factors. The interplay of up, down, and strange quarks within the proton is explored, shedding light on its internal structure. The contributions of strange quarks play a crucial role in the $D$-term and cosmological constants. We find that the flavor blindness of the isovector $D$-term form factor is only valid in flavor SU(3) symmetry., Comment: 6 pages, 2 figures

Motivated by the connection between pole-skipping phenomena of two point functions and four point out-of-time-order correlators, we study the pole-skipping phenomena for rotating BTZ black holes. In particular, we investigate the effect of rotations on the pole-skipping point for various fields with spin $s = 1/2, 1, 2/3$, extending the previous research for $s=0, 2$. We derive an analytic full tower of the pole-skipping points of fermionic ($s=1/2$) and vector ($s=1$) fields by the exact holographic Green's functions. For the \textit{non-extremal} black hole, the leading pole-skipping frequency is $\omega_{\text{leading}}=2\pi i T_h {(s-1+\nu \Omega)}/{(1-\Omega^2)}$ where $T_h$ is the temperature, $\Omega$ the rotation, and $\nu:=(\Delta_+ - \Delta_-)/2$, the difference of conformal dimensions ($\Delta_{\pm}$). These are confirmed by another independent method: the near-horizon analysis. For the \textit{extremal} black hole, we find that the leading pole-skipping frequency can occur at $\omega_{\text{leading}}^{\text{extremal}}=-2\pi i T_R {(s+1)}$ only when $\nu = s+1$, where $T_R$ is the temperature of the right moving mode. It is non-trivial because it cannot be achieved by simply taking the extreme limit ($T_h\rightarrow 0\,, \Omega\rightarrow 1$) of the non-extremal black hole result., Comment: 41 pages

In this work, we investigate spectral complexity and Krylov complexity in quantum billiard systems at finite temperature. We study both circle and stadium billiards as paradigmatic examples of integrable and non-integrable quantum-mechanical systems, respectively. We show that the saturation value and time scale of spectral complexity may be used to probe the non-integrability of the system since we find that when computed for the circle billiard, it saturates at a later time scale compared to the stadium billiards. This observation is verified for different temperatures. Furthermore, we study the Krylov complexity of the position operator and its associated Lanczos coefficients at finite temperature using the Wightman inner product. We find that the growth rate of the Lanczos coefficients saturates the conjectured universal bound at low temperatures. Additionally, we also find that even a subset of the Lanczos coefficients can potentially serve as an indicator of integrability, as they demonstrate erratic behavior specifically in the circle billiard case, in contrast to the stadium billiard. Finally, we also study Krylov entropy and verify its early-time logarithmic relation with Krylov complexity in both types of billiard systems., Comment: 31 pages, 17 figures

We describe varieties of minimal rational tangents on the wonderful symmetric varieties by marked Dynkin diagrams. An irreducible component of a variety of minimal rational tangents is a rational homogeneous space, and hence, we have a corresponding marked Dynkin diagram expression. On the other hand, we have marked Dynkin diagrams from the marked Kac diagram of a irreducible symmetric space by marking adjacent nodes to the marked node. We note that the above two diagrams coincide when the restricted root system is not of type A., Comment: 12 page, 8 tables

Earth's inner core (IC) serves as a reservoir for volatile elements, which significantly affects its behavior and properties. Recent studies suggest that superionicity can be observed in ice and iron hydrides under high-pressure and temperature conditions, providing an alternative understanding of the planet's interior. In this study, we demonstrated that electride formation drives the superionic state in iron hydride under IC pressure conditions. The electride stabilizes the iron lattice and provides a pathway for volatile diffusion. The coupling between lattice stability and superionicity is triggered near 100 GPa and enhanced at higher pressures. The electride-driven superionicity can also be generalized for volatiles in other rocky planetary cores. These findings provide new insights into the mechanisms of core formation and evolution of rocky planets., Comment: 26 pages, 4 main figures and 6 supplementary figures

In this paper, we address the performance degradation of efficient diffusion models by introducing Multi-architecturE Multi-Expert diffusion models (MEME). We identify the need for tailored operations at different time-steps in diffusion processes and leverage this insight to create compact yet high-performing models. MEME assigns distinct architectures to different time-step intervals, balancing convolution and self-attention operations based on observed frequency characteristics. We also introduce a soft interval assignment strategy for comprehensive training. Empirically, MEME operates 3.3 times faster than baselines while improving image generation quality (FID scores) by 0.62 (FFHQ) and 0.37 (CelebA). Though we validate the effectiveness of assigning more optimal architecture per time-step, where efficient models outperform the larger models, we argue that MEME opens a new design choice for diffusion models that can be easily applied in other scenarios, such as large multi-expert models., Comment: To be published in the AAAI 2024 Proceedings Main Track

We present the results of a radio multi-frequency ($\rm 3-340~GHz$) study of the blazar 3C~454.3. After subtracting the quiescent spectrum corresponding to optically thin emission, we found two individual synchrotron self-absorption (SSA) features in the wide-band spectrum. The one SSA had a relatively low turnover frequency ($\nu_{\rm m}$) in the range of $\rm 3-37~GHz$ (lower $\nu_{\rm m}$ SSA spectrum, LSS), and the other one had a relatively high $\nu_{\rm m}$ of $\rm 55-124~GHz$ (higher $\nu_{\rm m}$ SSA spectrum, HSS). Using the SSA parameters, we estimated magnetic field strengths at the surface where optical depth $\tau=1$. The estimated magnetic field strengths were $\rm >7~mG$ and $\rm >0.2~mG$ for the LSS and HSS, respectively. The LSS emitting region was magnetically dominated before the June 2014 $\gamma$-ray flare. The quasi-stationary component (C), $\sim 0.6~{\rm mas}$ apart from the 43 GHz radio core, became brighter than the core with decreasing observing frequency, and we found that component C was related to the LSS. A decrease in jet width was found near component C. As a moving component, K14 approached component C, and the flux density of the component was enhanced while the angular size decreased. The high intrinsic brightness temperature in the fluid frame was obtained as $T_{\rm B, int} \approx (7.0\pm1.0) \times 10^{11}~{\rm K}$ from the jet component after the 2015 August $\gamma$-ray flare, suggesting that component C is a high-energy emitting region. The observed local minimum of jet width and re-brightening behavior suggest a possible recollimation shock in component C., Comment: 16 pages, 12 figures, 5 tables, Accepted for publication in MNRAS

Self-supervised contrastive learning (CL) has achieved state-of-the-art performance in representation learning by minimizing the distance between positive pairs while maximizing that of negative ones. Recently, it has been verified that the model learns better representation with diversely augmented positive pairs because they enable the model to be more view-invariant. However, only a few studies on CL have considered the difference between augmented views, and have not gone beyond the hand-crafted findings. In this paper, we first observe that the score-matching function can measure how much data has changed from the original through augmentation. With the observed property, every pair in CL can be weighted adaptively by the difference of score values, resulting in boosting the performance of the existing CL method. We show the generality of our method, referred to as ScoreCL, by consistently improving various CL methods, SimCLR, SimSiam, W-MSE, and VICReg, up to 3%p in k-NN evaluation on CIFAR-10, CIFAR-100, and ImageNet-100. Moreover, we have conducted exhaustive experiments and ablations, including results on diverse downstream tasks, comparison with possible baselines, and improvement when used with other proposed augmentation methods. We hope our exploration will inspire more research in exploiting the score matching for CL.