arXiv subjects

Chen Xie

Publications and source records attributed to Chen Xie.

At least 37 records · Page 2Linked to original sources

arXiv2025

BEAM: Brainwave Empathy Assessment Model for Early Childhood

Empathy in young children is crucial for their social and emotional development, yet predicting it remains challenging. Traditional methods often only rely on self-reports or observer-based labeling, which are susceptible to bias and fail to objectively capture the process of empathy formation. EEG offers an objective alternative; however, current approaches primarily extract static patterns, neglecting temporal dynamics. To overcome these limitations, we propose a novel deep learning framework, the Brainwave Empathy Assessment Model (BEAM), to predict empathy levels in children aged 4-6 years. BEAM leverages multi-view EEG signals to capture both cognitive and emotional dimensions of empathy. The framework comprises three key components: 1) a LaBraM-based encoder for effective spatio-temporal feature extraction, 2) a feature fusion module to integrate complementary information from multi-view signals, and 3) a contrastive learning module to enhance class separation. Validated on the CBCP dataset, BEAM outperforms state-of-the-art methods across multiple metrics, demonstrating its potential for objective empathy assessment and providing a preliminary insight into early interventions in children's prosocial development.

cs.LG↗

arXiv2025

GTransPDM: A Graph-embedded Transformer with Positional Decoupling for Pedestrian Crossing Intention Prediction

Understanding and predicting pedestrian crossing behavioral intention is crucial for the driving safety of autonomous vehicles. Nonetheless, challenges emerge when using promising images or environmental context masks to extract various factors for time-series network modeling, causing pre-processing errors or a loss of efficiency. Typically, pedestrian positions captured by onboard cameras are often distorted and do not accurately reflect their actual movements. To address these issues, GTransPDM -- a Graph-embedded Transformer with a Position Decoupling Module -- was developed for pedestrian crossing intention prediction by leveraging multi-modal features. First, a positional decoupling module was proposed to decompose pedestrian lateral motion and encode depth cues in the image view. Then, a graph-embedded Transformer was designed to capture the spatio-temporal dynamics of human pose skeletons, integrating essential factors such as position, skeleton, and ego-vehicle motion. Experimental results indicate that the proposed method achieves 92% accuracy on the PIE dataset and 87% accuracy on the JAAD dataset, with a processing speed of 0.05ms. It outperforms the state-of-the-art in comparison.

cs.CV↗

arXiv2025

EdgePoint2: Compact Descriptors for Superior Efficiency and Accuracy

The field of keypoint extraction, which is essential for vision applications like Structure from Motion (SfM) and Simultaneous Localization and Mapping (SLAM), has evolved from relying on handcrafted methods to leveraging deep learning techniques. While deep learning approaches have significantly improved performance, they often incur substantial computational costs, limiting their deployment in real-time edge applications. Efforts to create lightweight neural networks have seen some success, yet they often result in trade-offs between efficiency and accuracy. Additionally, the high-dimensional descriptors generated by these networks poses challenges for distributed applications requiring efficient communication and coordination, highlighting the need for compact yet competitively accurate descriptors. In this paper, we present EdgePoint2, a series of lightweight keypoint detection and description neural networks specifically tailored for edge computing applications on embedded system. The network architecture is optimized for efficiency without sacrificing accuracy. To train compact descriptors, we introduce a combination of Orthogonal Procrustes loss and similarity loss, which can serve as a general approach for hypersphere embedding distillation tasks. Additionally, we offer 14 sub-models to satisfy diverse application requirements. Our experiments demonstrate that EdgePoint2 consistently achieves state-of-the-art (SOTA) accuracy and efficiency across various challenging scenarios while employing lower-dimensional descriptors (32/48/64). Beyond its accuracy, EdgePoint2 offers significant advantages in flexibility, robustness, and versatility. Consequently, EdgePoint2 emerges as a highly competitive option for visual tasks, especially in contexts demanding adaptability to diverse computational and communication constraints.

cs.CV↗

arXiv2025

Empowering AI to Generate Better AI Code: Guided Generation of Deep Learning Projects with LLMs

While large language models (LLMs) have been widely applied to code generation, they struggle with generating entire deep learning projects, which are characterized by complex structures, longer functions, and stronger reliance on domain knowledge than general-purpose code. An open-domain LLM often lacks coherent contextual guidance and domain expertise for specific projects, making it challenging to produce complete code that fully meets user requirements. In this paper, we propose a novel planning-guided code generation method, DLCodeGen, tailored for generating deep learning projects. DLCodeGen predicts a structured solution plan, offering global guidance for LLMs to generate the project. The generated plan is then leveraged to retrieve semantically analogous code samples and subsequently abstract a code template. To effectively integrate these multiple retrieval-augmented techniques, a comparative learning mechanism is designed to generate the final code. We validate the effectiveness of our approach on a dataset we build for deep learning code generation. Experimental results demonstrate that DLCodeGen outperforms other baselines, achieving improvements of 9.7% in CodeBLEU and 3.6% in human evaluation metrics.

cs.SE↗

arXiv2024

Disk Evolution Study Through Imaging of Nearby Young Stars (DESTINYS): Dynamical Evidence of a Spiral-Arm-Driving and Gap-Opening Protoplanet from SAO 206462 Spiral Motion

In the early stages of planetary system formation, young exoplanets gravitationally interact with their surrounding environments and leave observable signatures on protoplanetary disks. Among these structures, a pair of nearly symmetric spiral arms can be driven by a giant protoplanet. For the double-spiraled SAO 206462 protoplanetary disk, we obtained three epochs of observations spanning 7 yr using the Very Large Telescope's SPHERE instrument in near-infrared $J$-band polarized light. By jointly measuring the motion of the two spirals at three epochs, we obtained a rotation rate of $-0.85^\circ\pm0.05^\circ~{\rm yr}^{-1}$. This rate corresponds to a protoplanet at $66\pm3$ au on a circular orbit dynamically driving both spirals. The derived location agrees with the gap in ALMA dust-continuum observations, indicating that the spiral driver may also carve the observed gap. What is more, a dust filament at $\sim$63 au observed by ALMA coincides with the predicted orbit of the spiral-arm-driving protoplanet. This double-spiraled system is an ideal target for protoplanet imaging.

astro-ph.EP↗

arXiv2024

Language-Driven Interactive Traffic Trajectory Generation

Realistic trajectory generation with natural language control is pivotal for advancing autonomous vehicle technology. However, previous methods focus on individual traffic participant trajectory generation, thus failing to account for the complexity of interactive traffic dynamics. In this work, we propose InteractTraj, the first language-driven traffic trajectory generator that can generate interactive traffic trajectories. InteractTraj interprets abstract trajectory descriptions into concrete formatted interaction-aware numerical codes and learns a mapping between these formatted codes and the final interactive trajectories. To interpret language descriptions, we propose a language-to-code encoder with a novel interaction-aware encoding strategy. To produce interactive traffic trajectories, we propose a code-to-trajectory decoder with interaction-aware feature aggregation that synergizes vehicle interactions with the environmental map and the vehicle moves. Extensive experiments show our method demonstrates superior performance over previous SoTA methods, offering a more realistic generation of interactive traffic trajectories with high controllability via diverse natural language commands. Our code is available at https://github.com/X1a-jk/InteractTraj.git

cs.AI↗

arXiv2024

Multi-band reflectance and shadowing of RX J1604.3-2130 protoplanetary disk in scattered light

Context.Spatially-resoved cicrumstellar disk spectrum and composition can provide valuable insights into the bulk composition of forming planets, as well as the mineralogical signatures that emerge during and after planet formation. Aims. We aim to systemically extract the RX~J1604.3-213010 (J1604 hereafter) protoplanetary disk in high-contrast imaging observations, and obtain its multi-band reflectance in visible to near-infrared wavelengths. Methods. We obtained coronagraphic observations of J1604 from the Keck Observatory's NIRC2 instrument, and archival data from the Very Large Telescope's SPHERE instrument. Using archival images to remove star light and speckles, we recovered the J1604 disk and obtained its surface brightness using forward modeling. Together with polarization data, we obtained the relative reflectance of the disk in $R$, $J$, $H$ ($H2$ and $H3$), $K$ ($K1$ and $K2$), and $L'$ bands spanning two years. Results. Relative to the J1604 star, the resolved disk has a reflectance of ${\sim}10^{-1}$~arcsec$^{-2}$ in $R$ through $H$ bands and ${\sim}10^{-2}$~arcsec$^{-2}$ in $K$ and $L'$ bands, showing a blue color. Together with other systems, we summarized the multi-band reflectance for 9 systems. We also identified varying disk geometry structure, and a shadow that vanished between June and August in 2015. Conclusions. Motivated by broad-band observations, the deployment of cutting-edge technologies could yield higher-resolution reflection spectra, thereby informing the dust composition of disks in scattered light in the future. With multi-epoch observations, variable shadows have the potential to deepen insights into the dynamic characteristics of inner disk regions.

astro-ph.EP↗

arXiv2024

HW-SW Optimization of DNNs for Privacy-preserving People Counting on Low-resolution Infrared Arrays

Low-resolution infrared (IR) array sensors enable people counting applications such as monitoring the occupancy of spaces and people flows while preserving privacy and minimizing energy consumption. Deep Neural Networks (DNNs) have been shown to be well-suited to process these sensor data in an accurate and efficient manner. Nevertheless, the space of DNNs' architectures is huge and its manual exploration is burdensome and often leads to sub-optimal solutions. To overcome this problem, in this work, we propose a highly automated full-stack optimization flow for DNNs that goes from neural architecture search, mixed-precision quantization, and post-processing, down to the realization of a new smart sensor prototype, including a Microcontroller with a customized instruction set. Integrating these cross-layer optimizations, we obtain a large set of Pareto-optimal solutions in the 3D-space of energy, memory, and accuracy. Deploying such solutions on our hardware platform, we improve the state-of-the-art achieving up to 4.2x model size reduction, 23.8x code size reduction, and 15.38x energy reduction at iso-accuracy.

cs.LG↗

arXiv2023

A Companion in V1247 Ori Supported by Spiral Arm Pattern Motion

While there have been nearly two dozen of spiral arms detected from planet-forming disks in near-infrared scattered light, none of their substellar drivers have been confirmed. By observing spiral systems in at least two epochs spanning multiple years, and measuring the motion of the spirals, we can distinguish the cause of the spirals, and locate the orbits of the driving planets if they trigger the spirals. Upon a recent validation of this approach using the co-motion between a stellar companion and a spiral, we obtained a second epoch observation for the spiral system in the disk of V1247 Ori in the $H$-band polarized scattered light using VLT/SPHERE/IRDIS. Combining our observations with archival IRDIS data, we established a $4.8$ yr timeline to constrain the V1247 Ori spiral motion. We obtained a pattern speed of $0.40^{\circ} \pm 0.09^{\circ}$ yr$^{-1}$ for the north-east spiral. This corresponds to an orbital period of $900\pm200$ yr, and thus the semi-major axis of the hidden planetary driver is $118\pm19$ au for a 2.0 $\pm$ 0.1 M$_\odot$ central star. The location agrees with the gap in ALMA dust continuum observations, providing joint support for the existence of a companion driving the scattered-light spirals while carving a millimeter gap. With an angular separation of 0.29" $\pm$ 0.05", this hidden companion is an ideal target for JWST imaging.

astro-ph.EP↗

arXiv2023

Efficient Deep Learning Models for Privacy-preserving People Counting on Low-resolution Infrared Arrays

Ultra-low-resolution Infrared (IR) array sensors offer a low-cost, energy-efficient, and privacy-preserving solution for people counting, with applications such as occupancy monitoring. Previous work has shown that Deep Learning (DL) can yield superior performance on this task. However, the literature was missing an extensive comparative analysis of various efficient DL architectures for IR array-based people counting, that considers not only their accuracy, but also the cost of deploying them on memory- and energy-constrained Internet of Things (IoT) edge nodes. In this work, we address this need by comparing 6 different DL architectures on a novel dataset composed of IR images collected from a commercial 8x8 array, which we made openly available. With a wide architectural exploration of each model type, we obtain a rich set of Pareto-optimal solutions, spanning cross-validated balanced accuracy scores in the 55.70-82.70% range. When deployed on a commercial Microcontroller (MCU) by STMicroelectronics, the STM32L4A6ZG, these models occupy 0.41-9.28kB of memory, and require 1.10-7.74ms per inference, while consuming 17.18-120.43 $μ$J of energy. Our models are significantly more accurate than a previous deterministic method (up to +39.9%), while being up to 3.53x faster and more energy efficient. Further, our models' accuracy is comparable to state-of-the-art DL solutions on similar resolution sensors, despite a much lower complexity. All our models enable continuous, real-time inference on a MCU-based IoT node, with years of autonomous operation without battery recharging.

cs.CV↗

arXiv2023

The first scattered light images of HD 112810, a faint debris disk in the Sco-Cen association

Context: Circumstellar debris disks provide insight into the formation and early evolution of planetary systems. Resolved belts in particular help to locate planetesimals in exosystems, and can hint at the presence of disk-sculpting exoplanets. Aims: We study the circumstellar environment of HD 112810 (HIP 63439), a mid-F type star in the Sco-Cen association with a significant infrared excess indicating the presence of a circumstellar debris disk. Methods: We collected five high-contrast observations of HD 112810 with VLT/SPHERE. We identified a debris disk in scattered light, and found that the debris signature is robust over a number of epochs and a variety of reduction techniques. We modelled the disk, accounting for self-subtraction and assuming that it is optically thin. Results: We find a single-belt debris disk, with a radius of 118$\pm$9au and an inclination angle of ${75.7}^{+1.1}_{-1.3}$$°$. This is in good agreement with the constraints from SED modelling and from a partially-resolved ALMA image of the system. No planets are detected, though planets below the detection limit ($\sim$2.6M$_\textrm{J}$ at a projected separation of 118au) could be present and could have contributed to sculpting the ring of debris. Conclusions: HD 112810 adds to the growing inventory of debris disks imaged in scattered light. The disk is faint, but the radius and the inclination of the disk are promising for follow-up studies of the dust properties.

astro-ph.EP↗

arXiv2023

Dynamical detection of a companion driving a spiral arm in a protoplanetary disk

Radio and near-infrared observations have observed dozens of protoplanetary disks that host spiral arm features. Numerical simulations have shown that companions may excite spiral density waves in protoplanetary disks via companion-disk interaction. However, the lack of direct observational evidence for spiral-driving companions poses challenges to current theories of companion-disk interaction. Here we report multi-epoch observations of the binary system HD 100453 with the Spectro-Polarimetric High-contrast Exoplanet REsearch (SPHERE) facility at the Very Large Telescope. By recovering the spiral features via robustly removing starlight contamination, we measure spiral motion across 4 yr to perform dynamical motion analyses. The spiral pattern motion is consistent with the orbital motion of the eccentric companion. With this first observational evidence of a companion driving a spiral arm among protoplanetary disks, we directly and dynamically confirm the long-standing theory on the origin of spiral features in protoplanetary disks. With the pattern motion of companion-driven spirals being independent of companion mass, here we establish a feasible way of searching for hidden spiral-arm-driving planets that are beyond the detection of existing ground-based high-contrast imagers.

astro-ph.EP↗

arXiv2023

Regulating nanoscale heat transfer with Janus nanoparticles

Janus nanoparticles (JNPs) with heterogeneous compositions or interfacial properties can exhibit directional heating upon external excitation, such as laser radiation and magnetic field. This directional heating may be harnessed for new nanotechnology and biomedical applications. Understanding thermal transport and temperature control with JNP heating is critical for these advances. Here, we developed a numerical framework to analyze the asymmetric thermal transport in JNP heating under photothermal stimulation. We found that JNP-induced temperature contrast, defined as the ratio of temperature increase in the surrounding water, shows a substantial size and polar angle dependence. Notably, we discovered a significant enhancement of the temperature contrast under pulsed heating due to thermal confinement, compared with the continuous heating. This work brings new insights into the thermal responses of JNP heating and advances the field.

physics.app-ph↗

arXiv2022

Human Activity Recognition on Microcontrollers with Quantized and Adaptive Deep Neural Networks

Human Activity Recognition (HAR) based on inertial data is an increasingly diffused task on embedded devices, from smartphones to ultra low-power sensors. Due to the high computational complexity of deep learning models, most embedded HAR systems are based on simple and not-so-accurate classic machine learning algorithms. This work bridges the gap between on-device HAR and deep learning, proposing a set of efficient one-dimensional Convolutional Neural Networks (CNNs) deployable on general purpose microcontrollers (MCUs). Our CNNs are obtained combining hyper-parameters optimization with sub-byte and mixed-precision quantization, to find good trade-offs between classification results and memory occupation. Moreover, we also leverage adaptive inference as an orthogonal optimization to tune the inference complexity at runtime based on the processed input, hence producing a more flexible HAR system. With experiments on four datasets, and targeting an ultra-low-power RISC-V MCU, we show that (i) We are able to obtain a rich set of Pareto-optimal CNNs for HAR, spanning more than 1 order of magnitude in terms of memory, latency and energy consumption; (ii) Thanks to adaptive inference, we can derive >20 runtime operating modes starting from a single CNN, differing by up to 10% in classification scores and by more than 3x in inference complexity, with a limited memory overhead; (iii) on three of the four benchmarks, we outperform all previous deep learning methods, reducing the memory occupation by more than 100x. The few methods that obtain better performance (both shallow and deep) are not compatible with MCU deployment. (iv) All our CNNs are compatible with real-time on-device HAR with an inference latency <16ms. Their memory occupation varies in 0.05-23.17 kB, and their energy consumption in 0.005 and 61.59 uJ, allowing years of continuous operation on a small battery supply.

cs.LG↗

arXiv2022

Reference-star differential imaging on SPHERE/IRDIS

Reference-star differential imaging (RDI) is a promising technique in high-contrast imaging that is thought to be more sensitive to exoplanets and disks than angular differential imaging (ADI) at short angular separations (i.e., <0.3"). However, it is unknown whether the performance of RDI on ground-based instruments can be improved by using all the archival data to optimize the subtraction of stellar contributions. We characterize the performance of RDI on SPHERE/IRDIS data in direct imaging of exoplanets and disks. We made use of all the archival data in H23 obtained by SPHERE/IRDIS in the past five years to build a master reference library and perform RDI. In the point-source detection, RDI can outperform ADI at small angular separations (<0.4") if the observing conditions are around the median conditions of our master reference library. On average, RDI has a gain of ~0.8 mag over ADI at 0.15" separation for observations under median conditions. We demonstrate that including more reference targets in the master reference library can indeed help to improve the performance of RDI. In disk imaging, RDI can reveal more disk features and provide a more robust recovery of the disk morphology. We resolve 33 disks in total intensity (19 planet-forming disks and 14 debris disks), and 4 of them can only be detected with RDI. Two disks are resolved in scattered light for the first time. Three disks are detected in total intensity for the first time. The master reference library we built in this work can be easily implemented into legacy or future SPHERE surveys to perform RDI, achieving better performance than that of ADI. To obtain optimal RDI gains over ADI, we recommend future observations be carried out under seeing conditions of 0.6"-0.8".

astro-ph.EP↗

arXiv2022

Adaptive Random Forests for Energy-Efficient Inference on Microcontrollers

Random Forests (RFs) are widely used Machine Learning models in low-power embedded devices, due to their hardware friendly operation and high accuracy on practically relevant tasks. The accuracy of a RF often increases with the number of internal weak learners (decision trees), but at the cost of a proportional increase in inference latency and energy consumption. Such costs can be mitigated considering that, in most applications, inputs are not all equally difficult to classify. Therefore, a large RF is often necessary only for (few) hard inputs, and wasteful for easier ones. In this work, we propose an early-stopping mechanism for RFs, which terminates the inference as soon as a high-enough classification confidence is reached, reducing the number of weak learners executed for easy inputs. The early-stopping confidence threshold can be controlled at runtime, in order to favor either energy saving or accuracy. We apply our method to three different embedded classification tasks, on a single-core RISC-V microcontroller, achieving an energy reduction from 38% to more than 90% with a drop of less than 0.5% in accuracy. We also show that our approach outperforms previous adaptive ML methods for RFs.

cs.LG↗

arXiv2022

Ultra-compact Binary Neural Networks for Human Activity Recognition on RISC-V Processors

Human Activity Recognition (HAR) is a relevant inference task in many mobile applications. State-of-the-art HAR at the edge is typically achieved with lightweight machine learning models such as decision trees and Random Forests (RFs), whereas deep learning is less common due to its high computational complexity. In this work, we propose a novel implementation of HAR based on deep neural networks, and precisely on Binary Neural Networks (BNNs), targeting low-power general purpose processors with a RISC-V instruction set. BNNs yield very small memory footprints and low inference complexity, thanks to the replacement of arithmetic operations with bit-wise ones. However, existing BNN implementations on general purpose processors impose constraints tailored to complex computer vision tasks, which result in over-parametrized models for simpler problems like HAR. Therefore, we also introduce a new BNN inference library, which targets ultra-compact models explicitly. With experiments on a single-core RISC-V processor, we show that BNNs trained on two HAR datasets obtain higher classification accuracy compared to a state-of-the-art baseline based on RFs. Furthermore, our BNN reaches the same accuracy of a RF with either less memory (up to 91%) or more energy-efficiency (up to 70%), depending on the complexity of the features extracted by the RF.

cs.LG↗

arXiv2022

Energy-efficient and Privacy-aware Social Distance Monitoring with Low-resolution Infrared Sensors and Adaptive Inference

Low-resolution infrared (IR) Sensors combined with machine learning (ML) can be leveraged to implement privacy-preserving social distance monitoring solutions in indoor spaces. However, the need of executing these applications on Internet of Things (IoT) edge nodes makes energy consumption critical. In this work, we propose an energy-efficient adaptive inference solution consisting of the cascade of a simple wake-up trigger and a 8-bit quantized Convolutional Neural Network (CNN), which is only invoked for difficult-to-classify frames. Deploying such adaptive system on a IoT Microcontroller, we show that, when processing the output of a 8x8 low-resolution IR sensor, we are able to reduce the energy consumption by 37-57% with respect to a static CNN-based approach, with an accuracy drop of less than 2% (83% balanced accuracy).

cs.LG↗