arXiv subjects

Pengfei Li

Publications and source records attributed to Pengfei Li.

At least 163 records · Page 9Linked to original sources

arXiv2024

MonoOcc: Digging into Monocular Semantic Occupancy Prediction

Monocular Semantic Occupancy Prediction aims to infer the complete 3D geometry and semantic information of scenes from only 2D images. It has garnered significant attention, particularly due to its potential to enhance the 3D perception of autonomous vehicles. However, existing methods rely on a complex cascaded framework with relatively limited information to restore 3D scenes, including a dependency on supervision solely on the whole network's output, single-frame input, and the utilization of a small backbone. These challenges, in turn, hinder the optimization of the framework and yield inferior prediction results, particularly concerning smaller and long-tailed objects. To address these issues, we propose MonoOcc. In particular, we (i) improve the monocular occupancy prediction framework by proposing an auxiliary semantic loss as supervision to the shallow layers of the framework and an image-conditioned cross-attention module to refine voxel features with visual clues, and (ii) employ a distillation module that transfers temporal information and richer knowledge from a larger image backbone to the monocular semantic occupancy prediction framework with low cost of hardware. With these advantages, our method yields state-of-the-art performance on the camera-based SemanticKITTI Scene Completion benchmark. Codes and models can be accessed at https://github.com/ucaszyp/MonoOcc

cs.CV↗

arXiv2024

ChatMusician: Understanding and Generating Music Intrinsically with LLM

While Large Language Models (LLMs) demonstrate impressive capabilities in text generation, we find that their ability has yet to be generalized to music, humanity's creative language. We introduce ChatMusician, an open-source LLM that integrates intrinsic musical abilities. It is based on continual pre-training and finetuning LLaMA2 on a text-compatible music representation, ABC notation, and the music is treated as a second language. ChatMusician can understand and generate music with a pure text tokenizer without any external multi-modal neural structures or tokenizers. Interestingly, endowing musical abilities does not harm language abilities, even achieving a slightly higher MMLU score. Our model is capable of composing well-structured, full-length music, conditioned on texts, chords, melodies, motifs, musical forms, etc, surpassing GPT-4 baseline. On our meticulously curated college-level music understanding benchmark, MusicTheoryBench, ChatMusician surpasses LLaMA2 and GPT-3.5 on zero-shot setting by a noticeable margin. Our work reveals that LLMs can be an excellent compressor for music, but there remains significant territory to be conquered. We release our 4B token music-language corpora MusicPile, the collected MusicTheoryBench, code, model and demo in GitHub.

cs.SD↗

arXiv2024

A Distinct Radial Acceleration Relation across Brightest Cluster Galaxies and Galaxy Clusters

Recent studies reveal a radial acceleration relation (RAR) in galaxies, which illustrates a tight empirical correlation connecting the observational acceleration and the baryonic acceleration with a characteristic acceleration scale. However, a distinct RAR has been revealed on BCG-cluster scales with a seventeen times larger acceleration scale by the gravitational lensing effect. In this work, we systematically explored the acceleration and mass correlations between dynamical and baryonic components in 50 Brightest Cluster Galaxies (BCGs). To investigate the dynamical RAR in BCGs, we derived their dynamical accelerations from the stellar kinematics using the Jeans equation through Abel inversion and adopted the baryonic mass from the SDSS photometry. We explored the spatially resolved kinematic profiles with the largest integral field spectroscopy (IFS) data mounted by the Mapping Nearby Galaxies at Apache Point Observatory (MaNGA) survey. Our results demonstrate that the dynamical RAR in BCGs is consistent with the lensing RAR on BCG-cluster scales as well as a larger acceleration scale. This finding may imply that BCGs and galaxy clusters have fundamental differences from field galaxies. We also find a mass correlation, but it is less tight than the acceleration correlation.

astro-ph.GA↗

arXiv2024

Anytime-Competitive Reinforcement Learning with Policy Prior

This paper studies the problem of Anytime-Competitive Markov Decision Process (A-CMDP). Existing works on Constrained Markov Decision Processes (CMDPs) aim to optimize the expected reward while constraining the expected cost over random dynamics, but the cost in a specific episode can still be unsatisfactorily high. In contrast, the goal of A-CMDP is to optimize the expected reward while guaranteeing a bounded cost in each round of any episode against a policy prior. We propose a new algorithm, called Anytime-Competitive Reinforcement Learning (ACRL), which provably guarantees the anytime cost constraints. The regret analysis shows the policy asymptotically matches the optimal reward achievable under the anytime competitive constraints. Experiments on the application of carbon-intelligent computing verify the reward performance and cost constraint guarantee of ACRL.

cs.LG↗

arXiv2024

Tunable vortex bound states in multiband CsV3Sb5-derived kagome superconductors

Vortices and bound states offer an effective means of comprehending the electronic properties of superconductors. Recently, surface dependent vortex core states have been observed in the newly discovered kagome superconductors CsV3Sb5. Although the spatial distribution of the sharp zero energy conductance peak appears similar to Majorana bound states arising from the superconducting Dirac surface states, its origin remains elusive. In this study, we present observations of tunable vortex bound states (VBSs) in two chemically doped kagome superconductors Cs(V1-xTrx)3Sb5 (Tr=Ta or Ti), using low temperature scanning tunneling microscopy/spectroscopy. The CsV3Sb5-derived kagome superconductors exhibit full gap pairing superconductivity accompanied by the absence of long range charge orders, in contrast to pristine CsV3Sb5. Zero energy conductance maps demonstrate a field-driven continuous reorientation transition of the vortex lattice, suggesting multiband superconductivity. The Ta doped CsV3Sb5 displays the conventional cross shaped spatial evolution of Caroli de Gennes Matricon bound states, while the Ti doped CsV3Sb5 exhibits a sharp, non split zero bias conductance peak (ZBCP) that persists over a long distance across the vortex. The spatial evolution of the non split ZBCP is robust against surface effects and external magnetic field but is related to the doping concentrations. Our study reveals the tunable VBSs in multiband chemically doped CsV3Sb5 system and offers fresh insights into previously reported Y shaped ZBCP in a non quantum limit condition at the surface of kagome superconductor.

cond-mat.supr-con↗

arXiv2024

Online Allocation with Replenishable Budgets: Worst Case and Beyond

This paper studies online resource allocation with replenishable budgets, where budgets can be replenished on top of the initial budget and an agent sequentially chooses online allocation decisions without violating the available budget constraint at each round. We propose a novel online algorithm, called OACP (Opportunistic Allocation with Conservative Pricing), that conservatively adjusts dual variables while opportunistically utilizing available resources. OACP achieves a bounded asymptotic competitive ratio in adversarial settings as the number of decision rounds T gets large. Importantly, the asymptotic competitive ratio of OACP is optimal in the absence of additional assumptions on budget replenishment. To further improve the competitive ratio, we make a mild assumption that there is budget replenishment every T^* >= 1 decision rounds and propose OACP+ to dynamically adjust the total budget assignment for online allocation. Next, we move beyond the worst-case and propose LA-OACP (Learning-Augmented OACP/OACP+), a novel learning-augmented algorithm for online allocation with replenishable budgets. We prove that LA-OACP can improve the average utility compared to OACP/OACP+ when the ML predictor is properly trained, while still offering worst-case utility guarantees when the ML predictions are arbitrarily wrong. Finally, we run simulation studies of sustainable AI inference powered by renewables, validating our analysis and demonstrating the empirical benefits of LA-OACP.

cs.GT↗

arXiv2023

Robust Learning for Smoothed Online Convex Optimization with Feedback Delay

We study a challenging form of Smoothed Online Convex Optimization, a.k.a. SOCO, including multi-step nonlinear switching costs and feedback delay. We propose a novel machine learning (ML) augmented online algorithm, Robustness-Constrained Learning (RCL), which combines untrusted ML predictions with a trusted expert online algorithm via constrained projection to robustify the ML prediction. Specifically,we prove that RCL is able to guarantee$(1+λ)$-competitiveness against any given expert for any$λ>0$, while also explicitly training the ML model in a robustification-aware manner to improve the average-case performance. Importantly,RCL is the first ML-augmented algorithm with a provable robustness guarantee in the case of multi-step switching cost and feedback delay.We demonstrate the improvement of RCL in both robustness and average performance using battery management for electrifying transportationas a case study.

cs.LG↗

arXiv2023

ALECE: An Attention-based Learned Cardinality Estimator for SPJ Queries on Dynamic Workloads (Extended)

For efficient query processing, DBMS query optimizers have for decades relied on delicate cardinality estimation methods. In this work, we propose an Attention-based LEarned Cardinality Estimator (ALECE for short) for SPJ queries. The core idea is to discover the implicit relationships between queries and underlying dynamic data using attention mechanisms in ALECE's two modules that are built on top of carefully designed featurizations for data and queries. In particular, from all attributes in the database, the data-encoder module obtains organic and learnable aggregations which implicitly represent correlations among the attributes, whereas the query-analyzer module builds a bridge between the query featurizations and the data aggregations to predict the query's cardinality. We experimentally evaluate ALECE on multiple dynamic workloads. The results show that ALECE enables PostgreSQL's optimizer to achieve nearly optimal performance, clearly outperforming its built-in cardinality estimator and other alternatives.

cs.DB↗

arXiv2023

Reputation-Based Federated Learning Defense to Mitigate Threats in EEG Signal Classification

This paper presents a reputation-based threat mitigation framework that defends potential security threats in electroencephalogram (EEG) signal classification during model aggregation of Federated Learning. While EEG signal analysis has attracted attention because of the emergence of brain-computer interface (BCI) technology, it is difficult to create efficient learning models for EEG analysis because of the distributed nature of EEG data and related privacy and security concerns. To address these challenges, the proposed defending framework leverages the Federated Learning paradigm to preserve privacy by collaborative model training with localized data from dispersed sources and introduces a reputation-based mechanism to mitigate the influence of data poisoning attacks and identify compromised participants. To assess the efficiency of the proposed reputation-based federated learning defense framework, data poisoning attacks based on the risk level of training data derived by Explainable Artificial Intelligence (XAI) techniques are conducted on both publicly available EEG signal datasets and the self-established EEG signal dataset. Experimental results on the poisoned datasets show that the proposed defense methodology performs well in EEG signal classification while reducing the risks associated with security threats.

cs.CR↗

arXiv2023

A Robust Adversary Detection-Deactivation Method for Metaverse-oriented Collaborative Deep Learning

Metaverse is trending to create a digital circumstance that can transfer the real world to an online platform supported by large quantities of real-time interactions. Pre-trained Artificial Intelligence (AI) models are demonstrating their increasing capability in aiding the metaverse to achieve an excellent response with negligible delay, and nowadays, many large models are collaboratively trained by various participants in a manner named collaborative deep learning (CDL). However, several security weaknesses can threaten the safety of the CDL training process, which might result in fatal attacks to either the pre-trained large model or the local sensitive data sets possessed by an individual entity. In CDL, malicious participants can hide within the major innocent and silently uploads deceptive parameters to degenerate the model performance, or they can abuse the downloaded parameters to construct a Generative Adversarial Network (GAN) to acquire the private information of others illegally. To compensate for these vulnerabilities, this paper proposes an adversary detection-deactivation method, which can limit and isolate the access of potential malicious participants, quarantine and disable the GAN-attack or harmful backpropagation of received threatening gradients. A detailed protection analysis has been conducted on a Multiview CDL case, and results show that the protocol can effectively prevent harmful access by heuristic manner analysis and can protect the existing model by swiftly checking received gradients using only one low-cost branch with an embedded firewall.

cs.CR↗

arXiv2023

3D Implicit Transporter for Temporally Consistent Keypoint Discovery

Keypoint-based representation has proven advantageous in various visual and robotic tasks. However, the existing 2D and 3D methods for detecting keypoints mainly rely on geometric consistency to achieve spatial alignment, neglecting temporal consistency. To address this issue, the Transporter method was introduced for 2D data, which reconstructs the target frame from the source frame to incorporate both spatial and temporal information. However, the direct application of the Transporter to 3D point clouds is infeasible due to their structural differences from 2D images. Thus, we propose the first 3D version of the Transporter, which leverages hybrid 3D representation, cross attention, and implicit reconstruction. We apply this new learning system on 3D articulated objects and nonrigid animals (humans and rodents) and show that learned keypoints are spatio-temporally consistent. Additionally, we propose a closed-loop control strategy that utilizes the learned keypoints for 3D object manipulation and demonstrate its superior performance. Codes are available at https://github.com/zhongcl-thu/3D-Implicit-Transporter.

cs.CV↗

arXiv2023

Correlated BCS wavefunction approach to unconventional superconductors

We propose a modified BCS wavefunction as the ground state of a correlated superconductor with the correlation specified between $k$ and $-k$ electrons in the reciprocal space. Owing to this correlation, low-energy excitations are not conventional BCS Bogoliubov quasiparticles. They display at least four poles in the Green's function and their particle-hole weights in the tunneling and photon-emission spectrum become asymmetric. The superfluid density or superfluid stiffness also deviates from the BCS predictions with finite paramagnetic terms inside the total diamagnetic response. Moreover, a $d$-wave correlated pairing state becomes robust against weak disorders. Hence, this state can explain many mysterious features observed in unconventional superconductors like cuprates.

cond-mat.supr-con↗

arXiv2023

A solvable model for symmetry-breaking phase transitions

Analytically solvable models are benchmarks in studies of phase transitions and pattern-forming bifurcations. Such models are known for phase transitions of the second kind in uniform media, but not for localized states (solitons), as integrable equations which produce solitons do not admit intrinsic transitions in them. We introduce a solvable model for symmetry-breaking phase transitions of both the first and second kinds (alias sub- and supercritical bifurcations) for solitons pinned to a combined linear-nonlinear double-well potential, represented by a symmetric pair of delta-functions. Both self-focusing and defocusing signs of the nonlinearity are considered. In the former case, exact solutions are produced for symmetric and asymmetric solitons. The solutions explicitly demonstrate a switch between the symmetry-breaking transitions of the first and second kinds (i.e., sub- and supercritical bifurcations, respectively). In the self-defocusing model, the solution demonstrates the transition of the second kind which breaks antisymmetry of the first excited state.

nlin.PS↗

arXiv2023

DQS3D: Densely-matched Quantization-aware Semi-supervised 3D Detection

In this paper, we study the problem of semi-supervised 3D object detection, which is of great importance considering the high annotation cost for cluttered 3D indoor scenes. We resort to the robust and principled framework of selfteaching, which has triggered notable progress for semisupervised learning recently. While this paradigm is natural for image-level or pixel-level prediction, adapting it to the detection problem is challenged by the issue of proposal matching. Prior methods are based upon two-stage pipelines, matching heuristically selected proposals generated in the first stage and resulting in spatially sparse training signals. In contrast, we propose the first semisupervised 3D detection algorithm that works in the singlestage manner and allows spatially dense training signals. A fundamental issue of this new design is the quantization error caused by point-to-voxel discretization, which inevitably leads to misalignment between two transformed views in the voxel domain. To this end, we derive and implement closed-form rules that compensate this misalignment onthe-fly. Our results are significant, e.g., promoting ScanNet mAP@0.5 from 35.2% to 48.5% using 20% annotation. Codes and data will be publicly available.

cs.CV↗

arXiv2023

Dynamical hotness, star formation quenching and growth of supermassive black holes

A stellar system is dynamically hot when its kinetic energy is dominated by random motion represented by the velocity dispersion $σ_{\rm hot} (M_*)$. We use MaNGA data to obtain inner and outer dispersion of a galaxy, $σ_{\rm in}$ and $σ_{\rm out}$, to characterize its dynamical status and study its connection with star formation quenching and the growth of supermassive black hole (SMBH). We divide galaxies into fully quenched (FQGs), partially quenched (PQGs) and fully star-forming (FSGs) populations, and identify quenched central cores (QCCs) in PQGs. The galaxy distribution in $σ_{\rm in}/σ_{\rm hot}$-$σ_{\rm out}/σ_{\rm hot}$ diagram is L-shaped, consisting of a horizontal sequence ($σ_{\rm out}/σ_{\rm hot}\sim0$) and a vertical sequence ($σ_{\rm in}/σ_{\rm hot}\sim1$). FQGs and QCCs are located at the top of vertical sequence, $σ_{\rm out}/σ_{\rm hot}\sim1$, therefore they are dynamically hot over their entire bodies. PQGs reside along vertical sequence, so they have hot center but cold outskirt. FSGs are diverse and can be found in both sequences. Galaxy structural properties, star formation and AGN activities make a transition along horizontal sequence at $\log(σ_{\rm in}/σ_{\rm hot})\sim-0.3$, and along vertical sequence at $\log(σ_{\rm out}/σ_{\rm hot})\sim-0.3$. The fractions of optical AGNs and barred galaxies increase rapidly in the first transition and decline rapidly in the second; radio galaxies are located at the top of vertical sequence. Our results demonstrate that star formation quenching and SMBH growth are effective only in dynamically hot systems. A simple model along this line can reproduce the observed SMBH scaling relations. We discuss how secular processes and strong interactions can make a system dynamically hot, and lead to the SMBH growth and star formation quenching.

astro-ph.GA↗

arXiv2023

Masked Vision and Language Pre-training with Unimodal and Multimodal Contrastive Losses for Medical Visual Question Answering

Medical visual question answering (VQA) is a challenging task that requires answering clinical questions of a given medical image, by taking consider of both visual and language information. However, due to the small scale of training data for medical VQA, pre-training fine-tuning paradigms have been a commonly used solution to improve model generalization performance. In this paper, we present a novel self-supervised approach that learns unimodal and multimodal feature representations of input images and text using medical image caption datasets, by leveraging both unimodal and multimodal contrastive losses, along with masked language modeling and image text matching as pretraining objectives. The pre-trained model is then transferred to downstream medical VQA tasks. The proposed approach achieves state-of-the-art (SOTA) performance on three publicly available medical VQA datasets with significant accuracy improvements of 2.2%, 14.7%, and 1.7% respectively. Besides, we conduct a comprehensive analysis to validate the effectiveness of different components of the approach and study different pre-training settings. Our codes and models are available at https://github.com/pengfeiliHEU/MUMC.

cs.CV↗

arXiv2023

One-dimensional Townes solitons in dual-core systems with localized coupling

The recent creation of Townes solitons (TSs) in binary Bose-Einstein condensates and experimental demonstration of spontaneous symmetry breaking (SSB) in solitons propagating in dual-core optical fibers draw renewed interest to the TS and SSB phenomenology in these and other settings. In particular, stabilization of TSs, which are always unstable in free space, is a relevant problem with various ramifications. We introduce a system which admits exact solutions for both TSs and SSB of solitons. It is based on a dual-core waveguide with quintic self-focusing and fused (localized) coupling between the cores. The respective system of coupled nonlinear Schroedinger equations gives rise to exact solutions for full families of symmetric solitons and asymmetric ones, which are produced by the supercritical SSB bifurcation (i.e., the symmetry-breaking phase transition of the second kind). Stability boundaries of asymmetric solitons are identified by dint of numerical methods. Unstable solitons spontaneously transform into robust moderately asymmetric breathers or strongly asymmetric states with small intrinsic oscillations. The setup can be used in the design of photonic devices operating in coupling and switching regimes.

nlin.PS↗

arXiv2023

Distance Duality Test: The Evolution of Radio Sources Mimics a Nonexpanding Universe

Distance duality relation (DDR) marks a fundamental difference between expanding and nonexpanding Universes, as an expanding metric causes angular diameter distance smaller than luminosity distance by an extra factor of $(1+z)$. Here we report a test of this relation using two independent samples of ultracompact radio sources observed at 2.29 GHz and 5.0 GHz. The test with radio sources involves only geometry, so it is independent of cosmological models. Since the observed radio luminosities systematically increase with redshift, we do not assume a constant source size. Instead, we start with assuming the intensive property, luminosity density, does not evolve with redshift and then infer its evolution from the resultant DDR. We make the same assumption for both samples, and find it results in the same angular size-redshift relation. Interestingly, the resultant DDR is fully consistent with a nonexpanding Universe. Imposing the DDR predicted by the expanding Universe, we infer the radio luminosity density evolves as $ρ_L\propto(1+z)^3$. However, the perfect agreement with a nonexpanding Universe under the assumption of constant luminosity densities poses a conspiracy and fine-tuning problem: the size and luminosity density of ultracompact radio sources evolve in the way that precisely mimics a nonexpanding Universe.

astro-ph.CO↗