arXiv subjects

Siyu Chen

Publications and source records attributed to Siyu Chen.

At least 145 records · Page 8Linked to original sources

arXiv2023

PyPose: A Library for Robot Learning with Physics-based Optimization

Deep learning has had remarkable success in robotic perception, but its data-centric nature suffers when it comes to generalizing to ever-changing environments. By contrast, physics-based optimization generalizes better, but it does not perform as well in complicated tasks due to the lack of high-level semantic information and reliance on manual parametric tuning. To take advantage of these two complementary worlds, we present PyPose: a robotics-oriented, PyTorch-based library that combines deep perceptual models with physics-based optimization. PyPose's architecture is tidy and well-organized, it has an imperative style interface and is efficient and user-friendly, making it easy to integrate into real-world robotic applications. Besides, it supports parallel computing of any order gradients of Lie groups and Lie algebras and $2^{\text{nd}}$-order optimizers, such as trust region methods. Experiments show that PyPose achieves more than $10\times$ speedup in computation compared to the state-of-the-art libraries. To boost future research, we provide concrete examples for several fields of robot learning, including SLAM, planning, control, and inertial navigation.

cs.RO↗

arXiv2023

A Unified Framework of Policy Learning for Contextual Bandit with Confounding Bias and Missing Observations

We study the offline contextual bandit problem, where we aim to acquire an optimal policy using observational data. However, this data usually contains two deficiencies: (i) some variables that confound actions are not observed, and (ii) missing observations exist in the collected data. Unobserved confounders lead to a confounding bias and missing observations cause bias and inefficiency problems. To overcome these challenges and learn the optimal policy from the observed dataset, we present a new algorithm called Causal-Adjusted Pessimistic (CAP) policy learning, which forms the reward function as the solution of an integral equation system, builds a confidence set, and greedily takes action with pessimism. With mild assumptions on the data, we develop an upper bound to the suboptimality of CAP for the offline contextual bandit problem.

cs.LG↗

arXiv2023

Alkaline vents recreated in two dimensions to study pH gradients, precipitation morphology and molecule accumulation

Alkaline vents (AV) are hypothesized to have been a setting for the emergence of life, by creating strong gradients across inorganic membranes within chimney structures. In the past, 3-dimensional chimney structures were formed under laboratory conditions, however, no in situ visualisation or testing of the gradients was possible. We develop a quasi-2-dimensional microfluidic model of alkaline vents that allows spatio-temporal visualisation of mineral precipitation in low volume experiments. Upon injection of an alkaline fluid into an acidic, iron-rich solution, we observe a diverse set of precipitation morphologies, mainly controlled by flow-rate and ion-concentration. Using microscope imaging and pH dependent dyes, we show that finger-like precipitates can facilitate formation and maintenance of microscale pH gradients and accumulation of dispersed particles in confined geometries. Our findings establish a model to investigate the potential of gradients across a semi-permeable boundary for early compartmentalisation, accumulation and chemical reactions at the origins of life.

physics.bio-ph↗

arXiv2023

DiP: Learning Discriminative Implicit Parts for Person Re-Identification

In person re-identification (ReID) tasks, many works explore the learning of part features to improve the performance over global image features. Existing methods explicitly extract part features by either using a hand-designed image division or keypoints obtained with external visual systems. In this work, we propose to learn Discriminative implicit Parts (DiPs) which are decoupled from explicit body parts. Therefore, DiPs can learn to extract any discriminative features that can benefit in distinguishing identities, which is beyond predefined body parts (such as accessories). Moreover, we propose a novel implicit position to give a geometric interpretation for each DiP. The implicit position can also serve as a learning signal to encourage DiPs to be more position-equivariant with the identity in the image. Lastly, an additional DiP weighting is introduced to handle the invisible or occluded situation and further improve the feature representation of DiPs. Extensive experiments show that the proposed method achieves state-of-the-art performance on multiple person ReID benchmarks.

cs.CV↗

arXiv2023

Retinal blood flow speed quantification at the capillary level using temporal autocorrelation fitting OCTA

Optical coherence tomography angiography (OCTA) can visualize vasculature structures, but provides limited information about the blood flow speeds. Here, we present a second generation variable interscan time analysis (VISTA) OCTA, which evaluates a quantitative surrogate marker for blood flow speed in vasculature. At the capillary level, spatially compiled OCTA and a simple temporal autocorrelation model, ρ(τ) = exp(-ατ), were used to evaluate a temporal autocorrelation decay constant, α, as the blood flow speed marker. A 600 kHz A-scan rate swept-source provides short interscan time OCTA and fine A-scan spacing acquisition, while maintaining multi mm2 field of views for human retinal imaging. We demonstrate the cardiac pulsatility and repeatability of α measured with VISTA. We show different α for different retinal capillary plexuses in healthy eyes and present representative VISTA OCTA of eyes with diabetic retinopathy.

eess.IV↗

arXiv2022

SoccerNet 2022 Challenges Results

The SoccerNet 2022 challenges were the second annual video understanding challenges organized by the SoccerNet team. In 2022, the challenges were composed of 6 vision-based tasks: (1) action spotting, focusing on retrieving action timestamps in long untrimmed videos, (2) replay grounding, focusing on retrieving the live moment of an action shown in a replay, (3) pitch localization, focusing on detecting line and goal part elements, (4) camera calibration, dedicated to retrieving the intrinsic and extrinsic camera parameters, (5) player re-identification, focusing on retrieving the same players across multiple views, and (6) multiple object tracking, focusing on tracking players and the ball through unedited video streams. Compared to last year's challenges, tasks (1-2) had their evaluation metrics redefined to consider tighter temporal accuracies, and tasks (3-6) were novel, including their underlying data and annotations. More information on the tasks, challenges and leaderboards are available on https://www.soccer-net.org. Baselines and development kits are available on https://github.com/SoccerNet.

cs.CV↗

arXiv2022

A Spatiotemporal Model for Precise and Efficient Fully-automatic 3D Motion Correction in OCT

Optical coherence tomography (OCT) is a micrometer-scale, volumetric imaging modality that has become a clinical standard in ophthalmology. OCT instruments image by raster-scanning a focused light spot across the retina, acquiring sequential cross-sectional images to generate volumetric data. Patient eye motion during the acquisition poses unique challenges: Non-rigid, discontinuous distortions can occur, leading to gaps in data and distorted topographic measurements. We present a new distortion model and a corresponding fully-automatic, reference-free optimization strategy for computational motion correction in orthogonally raster-scanned, retinal OCT volumes. Using a novel, domain-specific spatiotemporal parametrization of forward-warping displacements, eye motion can be corrected continuously for the first time. Parameter estimation with temporal regularization improves robustness and accuracy over previous spatial approaches. We correct each A-scan individually in 3D in a single mapping, including repeated acquisitions used in OCT angiography protocols. Specialized 3D forward image warping reduces median runtime to < 9 s, fast enough for clinical use. We present a quantitative evaluation on 18 subjects with ocular pathology and demonstrate accurate correction during microsaccades. Transverse correction is limited only by ocular tremor, whereas submicron repeatability is achieved axially (0.51 um median of medians), representing a dramatic improvement over previous work. This allows assessing longitudinal changes in focal retinal pathologies as a marker of disease progression or treatment response, and promises to enable multiple new capabilities such as supersampled/super-resolution volume reconstruction and analysis of pathological eye motion occuring in neurological diseases.

eess.IV↗

arXiv2022

Market Design for Tradable Mobility Credits

Tradable mobility credit (TMC) schemes are an approach to travel demand management that have received significant attention in recent years. This paper proposes and analyzes alternative market models for a TMC system -- focusing on market design aspects such as allocation/expiration of tokens, rules governing trading, transaction fees, and regulator intervention -- and develops a methodology to explicitly model the dis-aggregate behavior of individuals within the market. Extensive simulation experiments are conducted within a combined mode and departure time context for the morning commute problem to compare the performance of the alternative designs relative to congestion pricing and a no-control scenario. The simulation experiments employ a day-to-day assignment framework wherein transportation demand is modeled using a logit-mixture model with income effects and supply is modeled using a standard bottleneck model. The results indicate that small fixed transaction fees can effectively mitigate undesirable behavior in the market without a significant loss in efficiency (total welfare) whereas proportional transaction fees are less effective both in terms of efficiency and in avoiding undesirable market behavior. Further, an allocation of tokens in continuous time can be beneficial in dealing with non-recurrent events and avoiding concentrated trading activity. In the presence of income effects, despite small fixed transaction fees, the TMC system yields a marginally higher social welfare than congestion pricing while attaining revenue neutrality. Further, it is more robust in the presence of forecasting errors and non-recurrent events due to the adaptiveness of the market. Finally, as expected, the TMC scheme is more equitable (when revenues from congestion pricing are not redistributed) although it is not guaranteed to be Pareto-improving when tokens are distributed equally.

econ.GN↗

arXiv2022

The radiation emitted from axion dark matter in a homogeneous magnetic field, and possibilities for detection

We study the direct radiation excited by oscillating axion (or axion-like particle) dark matter in a homogenous magnetic field and its detection scheme. We concretely derive the analytical expression of the axion-induced radiated power for a cylindrical uniform magnetic field. In the long wave limit, the radiation power is proportional to the square of the B-field volume and the axion mass $m_a$, whereas it oscillate as approaching the short wave limit and the peak powers are proportional to the side area of the cylindrical magnetic field and $m_a^{-2}$. The maximum power locates at mass $m_a\sim\frac{3π}{4R}$ for fixed radius $R$. Based on this characteristic of the power, we discuss a scheme to detect the axions in the mass range $1-10^4$\,neV, where four detectors of different bandwidths surround the B-field. The expected sensitivity for $m_a\lesssim1\,μ$eV under typical-parameter values can far exceed the existing constraints.

hep-ph↗

arXiv2022

Nonuniform grids for Brillouin zone integration and interpolation

We present two developments for the numerical integration of a function over the Brillouin zone. First, we introduce a nonuniform grid, which we refer to as the Farey grid, that generalizes regular grids. Second, we introduce symmetry-adapted Voronoi tessellation, a general technique to assign weights to the points in an arbitrary grid. Combining these two developments, we propose a strategy to perform Brillouin zone integration and interpolation that provides a significant computational advantage compared to the usual approach based on regular uniform grids. We demonstrate our methodology in the context of first principles calculations with the study of Kohn anomalies in the phonon dispersions of graphene and MgB2, and in the evaluation of the electron-phonon driven renormalization of the band gaps of diamond and bismuthene. In the phonon calculations, we find speedups by a factor of 3 to 4 when using density functional perturbation theory, and by a factor of 6 to 7 when using finite differences in conjunction with supercells. As a result, the computational expense between density functional perturbation theory and finite differences becomes comparable. For electron-phonon coupling calculations we find even larger speedups. Finally, we also demonstrate that the Farey grid can be expressed as a combination of the widely used regular grids, which should facilitate the adoption of this methodology.

cond-mat.other↗

arXiv2022

Muon Collider Physics Summary

The perspective of designing muon colliders with high energy and luminosity, which is being investigated by the International Muon Collider Collaboration, has triggered a growing interest in their physics reach. We present a concise summary of the muon colliders potential to explore new physics, leveraging on the unique possibility of combining high available energy with very precise measurements.

hep-ph↗

arXiv2022

The physics case of a 3 TeV muon collider stage

In the path towards a muon collider with center of mass energy of 10 TeV or more, a stage at 3 TeV emerges as an appealing option. Reviewing the physics potential of such muon collider is the main purpose of this document. In order to outline the progression of the physics performances across the stages, a few sensitivity projections for higher energy are also presented. There are many opportunities for probing new physics at a 3 TeV muon collider. Some of them are in common with the extensively documented physics case of the CLIC 3 TeV energy stage, and include measuring the Higgs trilinear coupling and testing the possible composite nature of the Higgs boson and of the top quark at the 20 TeV scale. Other opportunities are unique of a 3 TeV muon collider, and stem from the fact that muons are collided rather than electrons. This is exemplified by studying the potential to explore the microscopic origin of the current $g$-2 and $B$-physics anomalies, which are both related with muons.

hep-ph↗

arXiv2022

X-Learner: Learning Cross Sources and Tasks for Universal Visual Representation

In computer vision, pre-training models based on largescale supervised learning have been proven effective over the past few years. However, existing works mostly focus on learning from individual task with single data source (e.g., ImageNet for classification or COCO for detection). This restricted form limits their generalizability and usability due to the lack of vast semantic information from various tasks and data sources. Here, we demonstrate that jointly learning from heterogeneous tasks and multiple data sources contributes to universal visual representation, leading to better transferring results of various downstream tasks. Thus, learning how to bridge the gaps among different tasks and data sources is the key, but it still remains an open question. In this work, we propose a representation learning framework called X-Learner, which learns the universal feature of multiple vision tasks supervised by various sources, with expansion and squeeze stage: 1) Expansion Stage: X-Learner learns the task-specific feature to alleviate task interference and enrich the representation by reconciliation layer. 2) Squeeze Stage: X-Learner condenses the model to a reasonable size and learns the universal and generalizable representation for various tasks transferring. Extensive experiments demonstrate that X-Learner achieves strong performance on different tasks without extra annotations, modalities and computational costs compared to existing representation learning methods. Notably, a single X-Learner model shows remarkable gains of 3.0%, 3.3% and 1.8% over current pretrained models on 12 downstream datasets for classification, object detection and semantic segmentation.

cs.CV↗

arXiv2022

INTERN: A New Learning Paradigm Towards General Vision

Enormous waves of technological innovations over the past several years, marked by the advances in AI technologies, are profoundly reshaping the industry and the society. However, down the road, a key challenge awaits us, that is, our capability of meeting rapidly-growing scenario-specific demands is severely limited by the cost of acquiring a commensurate amount of training data. This difficult situation is in essence due to limitations of the mainstream learning paradigm: we need to train a new model for each new scenario, based on a large quantity of well-annotated data and commonly from scratch. In tackling this fundamental problem, we move beyond and develop a new learning paradigm named INTERN. By learning with supervisory signals from multiple sources in multiple stages, the model being trained will develop strong generalizability. We evaluate our model on 26 well-known datasets that cover four categories of tasks in computer vision. In most cases, our models, adapted with only 10% of the training data in the target domain, outperform the counterparts trained with the full set of data, often by a significant margin. This is an important step towards a promising prospect where such a model with general vision capability can dramatically reduce our reliance on data, thus expediting the adoption of AI technologies. Furthermore, revolving around our new paradigm, we also introduce a new data system, a new architecture, and a new benchmark, which, together, form a general vision ecosystem to support its future development in an open and inclusive manner. See project website at https://opengvlab.shlab.org.cn .

cs.CV↗

arXiv2022

Learning from Radiation at a Very High Energy Lepton Collider

We study the potential of lepton collisions with about $10\text{ TeV}$ center of mass energy to probe Electroweak, Higgs and Top short-distance physics at the $100\text{ TeV}$ scale, pointing out the interplay with the long-distance ($100\text{ GeV}$) phenomenon of Electroweak radiation. On one hand, we find that sufficiently accurate theoretical predictions require the resummed inclusion of radiation effects, which we perform at the double logarithmic order. On the other hand, we notice that short-distance physics does influence the emission of Electroweak radiation. Therefore the investigation of the radiation pattern can enhance the sensitivity to new short-distance physical laws. We illustrate these aspects by studying Effective Field Theory contact interactions in di-fermion and di-boson production, and comparing cross-section measurements that require or that exclude the emission of massive Electroweak bosons. The combination of the two types of measurements is found to enhance the sensitivity to the new interactions. Based on these results, we perform sensitivity projections to Higgs and Top Compositeness and to minimal $Z'$ new physics scenarios at future muon colliders.

hep-ph↗

arXiv2022

Near Interference-Free Space-Time User Scheduling for MmWave Cellular Network

The highly directional beams applied in millimeter wave (mmWave) cellular networks make it possible to achieve near interference-free (NIF) transmission under judiciously designed space-time user scheduling, where the power of intra-/inter-cell interference between any two users is below a predefined threshold. In this paper, we investigate two aspects of the NIF space-time user scheduling in a multi-cell mmWave network with multi-RF-chain base stations. Firstly, given that each user has a requirement on the number of space-time resource elements, we study the NIF user scheduling problem to minimize the unfulfilled user requirements, so that the space-time resources can be utilized most efficiently and meanwhile all strong interferences are avoided. A near-optimal scheduling algorithm is proposed with performance close to the lower bound of unfulfilled requirements. Furthermore, we study the joint NIF user scheduling and power allocation problem to minimize the power consumption under the constraint of rate requirements. Based on our proposed NIF scheduling, an energy-efficient joint scheduling and power allocation scheme is designed with limited channel state information, which outperforms the existing independent set based schemes, and has near-optimal performance as well.

cs.IT↗

arXiv2022

Revisiting the constraints on primordial black hole abundance with the isotropic gamma ray background

We revisit the constraints on primordial black holes (PBHs) in the mass range $10^{13}-10^{18}$ g by comparing the 100\,keV-5\,GeV gamma-ray background with isotropic flux from PBH Hawking radiation (HR). We investigate three effects that may update the constraints on the PBH abundance; i) reliably calculating the secondary spectra of HR for energy below 5\,GeV, ii) the contributions to the measured isotropic flux from the Galactic PBH HR and that from annihilation radiation due to evaporated positrons, iii) inclusion of astrophysical background from gamma-ray sources. The conservative constraint is significantly improved by more than an order of magnitude at $2\times10^{16}$g$\lesssim M\lesssim 10^{17}$g over the past relevant work, where the effect ii is dominant. After further accounting for the astrophysical background, more than a tenfold improvement extends to a much wider mass range $10^{15}$g$\lesssim M\lesssim 2\times 10^{17}$g.

astro-ph.CO↗

arXiv2021

Direct band gap and strong Rashba effect in van der Waals heterostructures of InSe and Sb single layers

Van der Waals heterostructures formed by stacking different types of 2D materials are attracting increasing attention due to new emergent physical properties such as interlayer excitons. Recently synthesized atomically thin indium selenide (InSe) and antimony (Sb) individually exhibit interesting electronic properties such as high electron mobility in the former and high hole mobility in the latter. In this work, we present a first-principles investigation on the stability and electronic properties of ultrathin bilayer heterostructures composed of InSe and Sb single layers. The calculated electronic band structures reveal a direct band gap semiconducting nature of the InSe/Sb heterostructures independent of stacking pattern. Taking spin-orbit coupling into account, we find a large Rashba spin splitting at the bottom of conduction band, which originates from the atomic spin-orbit coupling with the symmetry breaking in the heterostructure. The strength of the Rashba spin splitting can be tuned by applying in-plane biaxial strain or an out-of-plane external electric field. The presence of large Rashba spin splitting together with a suitable band gap in InSe/Sb bilayer heterostructures make them promising candidates for spin field-effect transistor and optoelectronic device applications.

cond-mat.mes-hall↗