Search

In recent years, the optical 4f system has drawn much attention in building high-speed and ultra-low-power optical neural networks (ONNs). Most optical systems suffer from the misalignment of the optical devices during installment. The performance of ONN based on the optical 4f system (4f-ONN) is considered sensitive to the misalignment in the optical path introduced. In order to comprehensively investigate the influence caused by the misalignment, we proposed a method for estimating the performance of a 4f-ONN in response to various misalignment in the context of the image classification task.The misalignment in numerical simulation is estimated by manipulating the optical intensity distributions in the fourth focus plane in the 4f system. Followed by a series of physical experiments to validate the simulation results. Using our method to test the impact of misalignment of 4f system on the classification accuracy of two popular image classification datasets, MNIST and Quickdraw16. On both datasets, we found that the performances of 4f-ONN generally degraded dramatically as the positioning error increased. Different positioning error tolerance in the misalignment orientations was observed over the two datasets. Classification performance could be preserved by positioning errors up to 200 microns in a specific direction.

Graph Neural Networks (GNNs) are a promising deep learning approach for circumventing many real-world problems on graph-structured data. However, these models usually have at least one of four fundamental limitations: over-smoothing, over-fitting, difficult to train, and strong homophily assumption. For example, Simple Graph Convolution (SGC) is known to suffer from the first and fourth limitations. To tackle these limitations, we identify a set of key designs including (D1) dilated convolution, (D2) multi-channel learning, (D3) self-attention score, and (D4) sign factor to boost learning from different types (i.e. homophily and heterophily) and scales (i.e. small, medium, and large) of networks, and combine them into a graph neural network, GPNet, a simple and efficient one-layer model. We theoretically analyze the model and show that it can approximate various graph filters by adjusting the self-attention score and sign factor. Experiments show that GPNet consistently outperforms baselines in terms of average rank, average accuracy, complexity, and parameters on semi-supervised and full-supervised tasks, and achieves competitive performance compared to state-of-the-art model with inductive learning task., Comment: 15 pages, 15 figures

We make the first attempt to define S-transformations for CFT$_2$ as linear mappings from closed to open sector linear spaces. The definition is based on closed-open sector linear space isomorphisms and boundary condition completeness. Diagonal RCFTs can be applied to our definition straight-forwardly, while more classes of CFT$_2$ are expected to be applicable. An unconventional open sector sewing, not among open sector sewing introduced by Lewellen, rises naturally and generalizes the definition. Our geometrical approach, partially inspired by string field theory, reveals the relationship between algebraic information in CFT$_2$ and curvature on surfaces., Comment: Errors in the method to determine the S-transformation coefficients, because the characters for the surfaces are not simply the product of characters for cylinders. Errors in the linear spaces assigned to surfaces in string vertices regions.More concrete definition for open sector linear spaces also required

There has been a long debate on the stable phase of iron under the Earth's inner core conditions. Because of the solid-liquid coexistence at the inner core boundary, the thermodynamic stability of solid phases directly relates to their melting temperatures, which remains considerable uncertainty. In the present study, we utilized a semi-empirical potential fitted to high-temperature ab initio data to perform a thermodynamic integration from classical systems described by this potential to ab initio systems. This method provides a smooth path for thermodynamic integration and significantly reduces the uncertainty caused by the finite-size effect. Our results suggest the hcp phase is the stable phase of pure iron under the inner core conditions, while the free energy difference between the hcp and bcc phases is tiny, on the order of 10s meV/atom near the melting temperature., Comment: 4 figures

We have selected six sources (G209.55-19.68S2, G205.46-14.56S1$_{-}$A, G203.21-11.20W2, G191.90-11.21S, G205.46-14.56S3, and G206.93-16.61W2) from the Atacama Large Millimeter/submillimeter Array Survey of Orion Planck Galactic Cold Clumps (ALMASOP), in which these sources have been mapped in the CO (J=2-1), SiO (J=5-4), and C$^{18}$O (J=2-1) lines. These sources have high-velocity SiO jets surrounded by low-velocity CO outflows. The SiO jets consist of a chain of knots. These knots have been thought to be produced by semi-periodical variations in jet velocity. Therefore, we adopt a shock-forming model, which uses such variations to estimate the inclination angle and velocity of the jets. We also derive the inclination angle of the CO outflows using the wide-angle wind-driven shell model, and find it to be broadly consistent with that of the associated SiO jets. In addition, we apply this shock-forming model to another three protostellar sources with SiO jets in the literature -- HH 211, HH 212, and L1448C(N) -- and find that their inclination angle and jet velocity are consistent with those previously estimated from proper motion and radial velocity studies.

We introduce a new conservation-based approach for traffic coordination modeling and control in a network of interconnected roads (NOIR) with switching movement phase rotations at every NOIR junction. For modeling of traffic evolution, we first assume that the movement phase rotation is cyclic at every NOIR junction, but the duration of each movement phase can be arbitrarily commanded by traffic signals. Then, we propose a novel concurrent switching dynamics (CSD) with deterministic transitions among a finite number of states, representing the NOIR movement phases. We define the CSD control as a cyclic receding horizon optimization problem with periodic quadratic cost and constraints. More specifically, the cost is defined so that the traffic density is minimized and the boundary inflow is uniformly distributed over the boundary inlet roads, whereas the cost parameters are periodically changed with time. The constraints are linear and imposed by a trapezoidal fundamental diagram at every NOIR road so that traffic feasibility is assured and traffic over-saturation is avoided. The success of the proposed traffic boundary control is demonstrated by simulation of traffic congestion control in Downtown Phoenix.

We present a comprehensive study of the gas kinematics associated with density structures at different spatial scales in the filamentary infrared dark cloud, G034.43+00.24 (G34). This study makes use of the H13CO+ (1-0) molecular line data from the ALMA Three-millimeter Observations of Massive Star-forming regions (ATOMS) survey, which has spatial and velocity resolution of 0.04 pc and 0.2 km/s, respectively. Several tens of dendrogram structures have been extracted in the position-position-velocity space of H13CO+, which include 21 small-scale leaves and 20 larger-scale branches. Overall, their gas motions are supersonic but they exhibit the interesting behavior where leaves tend to be less dynamically supersonic than the branches. For the larger-scale, branch structures, the observed velocity-size relation (i.e., velocity variation/dispersion versus size) are seen to follow the Larson scaling exponent while the smaller-scale, leaf structures show a systematic deviation and display a steeper slope. We argue that the origin of the observed kinematics of the branch structures is likely to be a combination of turbulence and gravity-driven ordered gas flows. In comparison, gravity-driven chaotic gas motion is likely at the level of small-scale leaf structures. The results presented in our previous paper and this current follow-up study suggest that the main driving mechanism for mass accretion/inflow observed in G34 varies at different spatial scales. We therefore conclude that a scale-dependent combined effect of turbulence and gravity is essential to explain the star-formation processes in G34., Comment: 11 pages, 6 figures, and 1 table. To appear in MNRAS

To understand the nature of SiO emission, we conducted ACA observations of the SiO (2-1) lines toward 146 massive star-forming regions, as part of the ALMA Three-millimeter Observations of Massive Star-forming regions (ATOMS) survey. We detected SiO emission in 128 (87.7$\%$) sources and identified 171 SiO clumps, 105 of which are spatially separated from 3 mm continuum emission. A large amount of the SiO line profiles (60$\%$) are non-Gaussian. The velocity dispersion of the SiO lines ranges from 0.3 to 5.43 km s$^{-1}$. In 63 sources the SiO clumps are associated with H$_\rm{II}$ regions characterized by H40$\alpha$ emission. We find that 68$\%$ (116) of the SiO clumps are associated with strong outflows. The median velocity dispersion of the SiO line for outflow sources and non-outflow sources is 1.91 km s$^{-1}$ and 0.99 km s$^{-1}$, respectively. These results indicate that outflow activities could be connected to strongly shocked gas. The velocity dispersion and [SiO]/[H$^{13}$CO$^+$] intensity ratio do not show any correlation with the dust temperature and particle number density of clumps. We find a positive correlation between the SiO line luminosity and the bolometric luminosity, implying stronger shock activities are associated with more luminous proto-clusters. The SiO clumps in associations with H$_\rm{II}$ regions were found to show a steeper feature in $L_\rm{sio}$/$L_\rm{bol}$. The SiO line luminosity and the fraction of shocked gas have no apparent evidence of correlation with the evolutionary stages traced by luminosity to mass ratio ($L_\rm{bol}/M$)., Comment: 18 pages, 10 figures

We present new 3-mm continuum and molecular lines observations from the ATOMS survey towards the massive protostellar clump, MM1, located in the filamentary infrared dark cloud (IRDC), G034.43+00.24 (G34). The lines observed are the tracers of either dense gas (e.g. HCO+/H13CO+ J = 1-0) or outflows (e.g. CS J = 2-1). The most complete picture to date of seven cores in MM1 is revealed by dust continuum emission. These cores are found to be gravitationally bound, with virial parameter, $\alpha_{vir}<2$. At least four outflows are identified in MM1 with a total outflowing mass of $\sim 45 M_\odot$, and a total energy of $\sim 1\times 10^{47}$ erg, typical of outflows from a B0-type star. Evidence of hierarchical fragmentation, where turbulence dominates over thermal pressure, is observed at both the cloud and the clump scales. This could be linked to the scale-dependent, dynamical mass inflow/accretion on clump and core scales. We therefore suggest that the G34 cloud could be undergoing a dynamical mass inflow/accretion process linked to the multiscale fragmentation, which leads to the sequential formation of fragments of the initial cloud, clumps, and ultimately dense cores, the sites of star formation., Comment: 14 pages with 6 figures, and in press

To enhance human-robot social interaction, it is essential for robots to process multiple social cues in a complex real-world environment. However, incongruency of input information across modalities is inevitable and could be challenging for robots to process. To tackle this challenge, our study adopted the neurorobotic paradigm of crossmodal conflict resolution to make a robot express human-like social attention. A behavioural experiment was conducted on 37 participants for the human study. We designed a round-table meeting scenario with three animated avatars to improve ecological validity. Each avatar wore a medical mask to obscure the facial cues of the nose, mouth, and jaw. The central avatar shifted its eye gaze while the peripheral avatars generated sound. Gaze direction and sound locations were either spatially congruent or incongruent. We observed that the central avatar's dynamic gaze could trigger crossmodal social attention responses. In particular, human performances are better under the congruent audio-visual condition than the incongruent condition. Our saliency prediction model was trained to detect social cues, predict audio-visual saliency, and attend selectively for the robot study. After mounting the trained model on the iCub, the robot was exposed to laboratory conditions similar to the human experiment. While the human performances were overall superior, our trained model demonstrated that it could replicate attention responses similar to humans., Comment: accepted for publication in the International Journal of Social Robotics

Sensory and emotional experiences such as pain and empathy are essential for mental and physical health. Cognitive neuroscience has been working on revealing mechanisms underlying pain and empathy. Furthermore, as trending research areas, computational pain recognition and empathic artificial intelligence (AI) show progress and promise for healthcare or human-computer interaction. Although AI research has recently made it increasingly possible to create artificial systems with affective processing, most cognitive neuroscience and AI research do not jointly address the issues of empathy in AI and cognitive neuroscience. The main aim of this paper is to introduce key advances, cognitive challenges and technical barriers in computational pain recognition and the implementation of artificial empathy. Our discussion covers the following topics: How can AI recognize pain from unimodal and multimodal information? Is it crucial for AI to be empathic? What are the benefits and challenges of empathic AI? Despite some consensus on the importance of AI, including empathic recognition and responses, we also highlight future challenges for artificial empathy and possible paths from interdisciplinary perspectives. Furthermore, we discuss challenges for responsible evaluation of cognitive methods and computational techniques and show approaches to future work to contribute to affective assistants capable of empathy.

Filaments play an important role in star formation, but the formation process of filaments themselves is still unclear. The high-mass star forming clump G286.21+0.17 (G286 for short) that contains an "L" type filament was thought to undergo global collapse. Our high resolution ALMA band 3 observations resolve the gas kinematics of G286 and reveal two sub-clumps with very different velocities inside it. We find that the "blue profile" (an indicator of gas infall) of HCO+ lines in single dish observations of G286 is actually caused by gas emission from the two sub-clumps rather than gas infall. We advise great caution in interpreting gas kinematics (e.g., infall) from line profiles toward distant massive clumps in single dish observations. Energetic outflows are identified in G286 but the outflows are not strong enough to drive expansion of the two sub-clumps. The two parts of the "L" type filament ("NW-SE" and "NE-SW" filaments) show prominent velocity gradients perpendicular to their major axes, indicating that they are likely formed due to large-scale compression flows. We argue that the large-scale compression flows could be induced by the expansion of nearby giant HII regions. The "NW-SE" and "NE-SW" filaments seem to be in collision, and a large amount of gas has been accumulated in the junction region where the most massive core G286c1 forms., Comment: Accepted by MNRAS

As machine learning is applied more widely, it is necessary to have a machine learning platform for both infrastructure administrators and users including expert data scientists and citizen data scientists to improve their productivity. However, existing machine learning platforms are ill-equipped to address the "Machine Learning tech debts" such as glue code, reproducibility, and portability. Furthermore, existing platforms only take expert data scientists into consideration, and thus they are inflexible for infrastructure administrators and non-user-friendly for citizen data scientists. We propose Submarine, a unified machine learning platform, to address the challenges., Comment: 9 pages