arXiv subjects

Xinlei Chen

Publications and source records attributed to Xinlei Chen.

At least 109 records · Page 6Linked to original sources

arXiv2025

Test-Time Training on Video Streams

Prior work has established Test-Time Training (TTT) as a general framework to further improve a trained model at test time. Before making a prediction on each test instance, the model is first trained on the same instance using a self-supervised task such as reconstruction. We extend TTT to the streaming setting, where multiple test instances - video frames in our case - arrive in temporal order. Our extension is online TTT: The current model is initialized from the previous model, then trained on the current frame and a small window of frames immediately before. Online TTT significantly outperforms the fixed-model baseline for four tasks, on three real-world datasets. The improvements are more than 2.2x and 1.5x for instance and panoptic segmentation. Surprisingly, online TTT also outperforms its offline variant that accesses strictly more information, training on all frames from the entire test video regardless of temporal order. This finding challenges those in prior work using synthetic videos. We formalize a notion of locality as the advantage of online over offline TTT, and analyze its role with ablations and a theory based on bias-variance trade-off.

cs.CV↗

arXiv2024

MetaMorph: Multimodal Understanding and Generation via Instruction Tuning

In this work, we propose Visual-Predictive Instruction Tuning (VPiT) - a simple and effective extension to visual instruction tuning that enables a pretrained LLM to quickly morph into an unified autoregressive model capable of generating both text and visual tokens. VPiT teaches an LLM to predict discrete text tokens and continuous visual tokens from any input sequence of image and text data curated in an instruction-following format. Our empirical investigation reveals several intriguing properties of VPiT: (1) visual generation ability emerges as a natural byproduct of improved visual understanding, and can be unlocked efficiently with a small amount of generation data; (2) while we find understanding and generation to be mutually beneficial, understanding data contributes to both capabilities more effectively than generation data. Building upon these findings, we train our MetaMorph model and achieve competitive performance on both visual understanding and generation. In visual generation, MetaMorph can leverage the world knowledge and reasoning abilities gained from LLM pretraining, and overcome common failure modes exhibited by other generation models. Our results suggest that LLMs may have strong "prior" vision capabilities that can be efficiently adapted to both visual understanding and generation with a relatively simple instruction tuning process.

cs.CV↗

arXiv2024

Simultaneous Multiband Photometry of the Early Optical Afterglow of GRB 240825A with Mephisto

Gamma-ray bursts (GRBs) are the most luminous transients in the universe. The interaction of the relativistic jet with the circumburst medium produces an afterglow and generates multiwavelength emission. In this work, we present simultaneous multiband photometry of GRB~240825A with the Multi-channel Photometric Survey Telescope (Mephisto) and analyze its temporal and spectral properties. The measurement began 128 seconds after the GRB trigger and continued until the fourth day when the afterglow essentially diminished and the measured brightness was close to that of the host galaxy. Based on the multiband light curves in the $uvgriz$ bands, we find that the optical flux density satisfies $F_{ν,{\rm obs}}\propto t^{-1.34}ν^{-2.48}$ with a spectral index of $2.48$ much larger than those of most other GRBs. To reconcile the measured much softer spectral energy distribution (SED) with that predicted by the standard afterglow model, an extra host-galaxy extinction of $E_{B-V}\sim(0.37-0.57)$ mag is required. We interpreted this excess as arising from a dense circumburst medium. We further find that the SED of the optical afterglow hardened as the afterglow decayed and the color excess $E_{B-V}$ decreased $\sim0.26$ mag from 100 seconds to 3000 seconds after the GRB trigger. Finally, we analyze the properties of the host galaxy of GRB~240825A based on data from the SDSS, PanSTARRS and HSC-SSP surveys. For a host redshift of $z=0.659$, the stellar mass and star formation rate of the host galaxy are estimated to be $\log(M_*/M_\odot)=10.0^{+0.3}_{-0.3}$ and $\log({\rm SFR}/M_{\odot}{\rm yr}^{-1})= 0.6^{+0.8}_{-3.3}$, respectively, pointing to a gas-rich, star-forming, medium-size galaxy.

astro-ph.HE↗

arXiv2024

Learning Video Representations without Natural Videos

We show that useful video representations can be learned from synthetic videos and natural images, without incorporating natural videos in the training. We propose a progression of video datasets synthesized by simple generative processes, that model a growing set of natural video properties (e.g., motion, acceleration, and shape transformations). The downstream performance of video models pre-trained on these generated datasets gradually increases with the dataset progression. A VideoMAE model pre-trained on our synthetic videos closes 97.2\% of the performance gap on UCF101 action classification between training from scratch and self-supervised pre-training from natural videos, and outperforms the pre-trained model on HMDB51. Introducing crops of static images to the pre-training stage results in similar performance to UCF101 pre-training and outperforms the UCF101 pre-trained model on 11 out of 14 out-of-distribution datasets of UCF101-P. Analyzing the low-level properties of the datasets, we identify correlations between frame diversity, frame similarity to natural data, and downstream performance. Our approach provides a more controllable and transparent alternative to video data curation processes for pre-training.

cs.CV↗

arXiv2024

On the Surprising Effectiveness of Attention Transfer for Vision Transformers

Conventional wisdom suggests that pre-training Vision Transformers (ViT) improves downstream performance by learning useful representations. Is this actually true? We investigate this question and find that the features and representations learned during pre-training are not essential. Surprisingly, using only the attention patterns from pre-training (i.e., guiding how information flows between tokens) is sufficient for models to learn high quality features from scratch and achieve comparable downstream performance. We show this by introducing a simple method called attention transfer, where only the attention patterns from a pre-trained teacher ViT are transferred to a student, either by copying or distilling the attention maps. Since attention transfer lets the student learn its own features, ensembling it with a fine-tuned teacher also further improves accuracy on ImageNet. We systematically study various aspects of our findings on the sufficiency of attention maps, including distribution shift settings where they underperform fine-tuning. We hope our exploration provides a better understanding of what pre-training accomplishes and leads to a useful alternative to the standard practice of fine-tuning

cs.LG↗

arXiv2024

FedASTA: Federated adaptive spatial-temporal attention for traffic flow prediction

Mobile devices and the Internet of Things (IoT) devices nowadays generate a large amount of heterogeneous spatial-temporal data. It remains a challenging problem to model the spatial-temporal dynamics under privacy concern. Federated learning (FL) has been proposed as a framework to enable model training across distributed devices without sharing original data which reduce privacy concern. Personalized federated learning (PFL) methods further address data heterogenous problem. However, these methods don't consider natural spatial relations among nodes. For the sake of modeling spatial relations, Graph Neural Netowork (GNN) based FL approach have been proposed. But dynamic spatial-temporal relations among edge nodes are not taken into account. Several approaches model spatial-temporal dynamics in a centralized environment, while less effort has been made under federated setting. To overcome these challeges, we propose a novel Federated Adaptive Spatial-Temporal Attention (FedASTA) framework to model the dynamic spatial-temporal relations. On the client node, FedASTA extracts temporal relations and trend patterns from the decomposed terms of original time series. Then, on the server node, FedASTA utilize trend patterns from clients to construct adaptive temporal-spatial aware graph which captures dynamic correlation between clients. Besides, we design a masked spatial attention module with both static graph and constructed adaptive graph to model spatial dependencies among clients. Extensive experiments on five real-world public traffic flow datasets demonstrate that our method achieves state-of-art performance in federated scenario. In addition, the experiments made in centralized setting show the effectiveness of our novel adaptive graph construction approach compared with other popular dynamic spatial-temporal aware methods.

cs.LG↗

arXiv2024

EmbodiedCity: A Benchmark Platform for Embodied Agent in Real-world City Environment

Embodied artificial intelligence emphasizes the role of an agent's body in generating human-like behaviors. The recent efforts on EmbodiedAI pay a lot of attention to building up machine learning models to possess perceiving, planning, and acting abilities, thereby enabling real-time interaction with the world. However, most works focus on bounded indoor environments, such as navigation in a room or manipulating a device, with limited exploration of embodying the agents in open-world scenarios. That is, embodied intelligence in the open and outdoor environment is less explored, for which one potential reason is the lack of high-quality simulators, benchmarks, and datasets. To address it, in this paper, we construct a benchmark platform for embodied intelligence evaluation in real-world city environments. Specifically, we first construct a highly realistic 3D simulation environment based on the real buildings, roads, and other elements in a real city. In this environment, we combine historically collected data and simulation algorithms to conduct simulations of pedestrian and vehicle flows with high fidelity. Further, we designed a set of evaluation tasks covering different EmbodiedAI abilities. Moreover, we provide a complete set of input and output interfaces for access, enabling embodied agents to easily take task requirements and current environmental observations as input and then make decisions and obtain performance evaluations. On the one hand, it expands the capability of existing embodied intelligence to higher levels. On the other hand, it has a higher practical value in the real world and can support more potential applications for artificial general intelligence. Based on this platform, we evaluate some popular large language models for embodied intelligence capabilities of different dimensions and difficulties.

cs.AI↗

arXiv2024

Scaling Proprioceptive-Visual Learning with Heterogeneous Pre-trained Transformers

One of the roadblocks for training generalist robotic models today is heterogeneity. Previous robot learning methods often collect data to train with one specific embodiment for one task, which is expensive and prone to overfitting. This work studies the problem of learning policy representations through heterogeneous pre-training on robot data across different embodiments and tasks at scale. We propose Heterogeneous Pre-trained Transformers (HPT), which pre-train a large, shareable trunk of a policy neural network to learn a task and embodiment agnostic shared representation. This general architecture aligns the specific proprioception and vision inputs from distinct embodiments to a short sequence of tokens and then processes such tokens to map to control robots for different tasks. Leveraging the recent large-scale multi-embodiment real-world robotic datasets as well as simulation, deployed robots, and human video datasets, we investigate pre-training policies across heterogeneity. We conduct experiments to investigate the scaling behaviors of training objectives, to the extent of 52 datasets. HPTs outperform several baselines and enhance the fine-tuned policy performance by over 20% on unseen tasks in multiple simulator benchmarks and real-world settings. See the project website (https://liruiw.github.io/hpt/) for code and videos.

cs.RO↗

arXiv2024

Range-SLAM: Ultra-Wideband-Based Smoke-Resistant Real-Time Localization and Mapping

This paper presents Range-SLAM, a real-time, lightweight SLAM system designed to address the challenges of localization and mapping in environments with smoke and other harsh conditions using Ultra-Wideband (UWB) signals. While optical sensors like LiDAR and cameras struggle in low-visibility environments, UWB signals provide a robust alternative for real-time positioning. The proposed system uses general UWB devices to achieve accurate mapping and localization without relying on expensive LiDAR or other dedicated hardware. By utilizing only the distance and Received Signal Strength Indicator (RSSI) provided by UWB sensors in relation to anchors, we combine the motion of the tag-carrying agent with raycasting algorithm to construct a 2D occupancy grid map in real time. To enhance localization in challenging conditions, a Weighted Least Squares (WLS) method is employed. Extensive real-world experiments, including smoke-filled environments and simulated

cs.RO↗

arXiv2024

Palantir: Towards Efficient Super Resolution for Ultra-high-definition Live Streaming

Neural enhancement through super-resolution (SR) deep neural networks (DNNs) opens up new possibilities for ultra-high-definition (UHD) live streaming over existing encoding and networking infrastructure. Yet, the heavy SR DNN inference overhead leads to severe deployment challenges. To reduce the overhead, existing systems propose to apply DNN-based SR only on carefully selected anchor frames while upscaling non-anchor frames via the lightweight reusing-based SR approach. However, frame-level scheduling is coarse-grained and fails to deliver optimal efficiency. In this work, we propose Palantir, the first neural-enhanced UHD live streaming system with fine-grained patch-level scheduling. Two novel techniques are incorporated into Palantir to select the most beneficial anchor patches and support latency-sensitive UHD live streaming applications. Firstly, under the guidance of our pioneering and theoretical analysis, Palantir constructs a directed acyclic graph (DAG) for lightweight yet accurate SR quality estimation under any possible anchor patch set. Secondly, to further optimize the scheduling latency, Palantir improves parallelizability by refactoring the computation subprocedure of the estimation process into a sparse matrix-matrix multiplication operation. The evaluation results suggest that Palantir incurs a negligible scheduling latency accounting for less than 5.7% of the end-to-end latency requirement. When compared to the naive method of applying DNN-based SR on all the frames, Palantir can reduce the SR DNN inference overhead by 20 times (or 60 times) while preserving 54.0-82.6% (or 32.8-64.0%) of the quality gain. When compared to the state-of-the-art real-time frame-level scheduling strategy, Palantir can reduce the SR DNN inference overhead by 80.1% at most (and 38.4% on average) without sacrificing the video quality.

cs.MM↗

arXiv2024

Massive Activations in Large Language Models

We observe an empirical phenomenon in Large Language Models (LLMs) -- very few activations exhibit significantly larger values than others (e.g., 100,000 times larger). We call them massive activations. First, we demonstrate the widespread existence of massive activations across various LLMs and characterize their locations. Second, we find their values largely stay constant regardless of the input, and they function as indispensable bias terms in LLMs. Third, these massive activations lead to the concentration of attention probabilities to their corresponding tokens, and further, implicit bias terms in the self-attention output. Last, we also study massive activations in Vision Transformers. Code is available at https://github.com/locuslab/massive-activations.

cs.CL↗

arXiv2024

Black hole mass and optical radiation mechanism of the tidal disruption event AT 2023clx

We present the optical light curves of the tidal disruption event (TDE) AT 2023clx in the declining phase, observed with Mephisto. Combining our light curve with the ASAS-SN and ATLAS data in the rising phase, and fitting the composite multi-band light curves with MOSFiT, we estimate black hole mass of AT 2023clx is between $10^{5.67}$--$10^{5.82}~M_{\odot}$. This event may be caused by either a full disruption of a $0.1~M_{\odot}$ star, or a partial disruption of a $0.99~M_{\odot}$ star, depending on the data adopted for the rising phase. Based on those fit results and the non-detection of soft X-ray photons in the first 90 days, we propose that the observed optical radiation is powered by stream-stream collision. We speculate that the soft X-ray photons may gradually emerge in 100--600 days after the optical peak, when the debris is fully circularized into a compact accretion disk.

astro-ph.HE↗

arXiv2024

Early-phase simultaneous multiband observations of the Type II supernova SN 2024ggi with Mephisto

We present early-phase good-cadence (hour-to-day) simultaneous multiband ($ugi$ and $vrz$ bands) imaging of the nearby supernova SN~2024ggi, which exploded in the nearby galaxy, NGC 3621. A quick follow-up was conducted within less than a day after the explosion and continued $\sim$23 days. The $uvg$ band light curves display a rapid rise ($\sim$1.4 mag day$^{-1}$) to maximum in $\sim$4 days and absolute magnitude $M_{g}\sim$--17.75 mag. The post-peak decay rate in redder bands is $\sim$0.01 mag day$^{-1}$. Different colors (e.g., $u-g$ and $v-r$) of SN~2024ggi are slightly redder than SN 2023ixf. A significant rise ($\sim$12.5 kK) in black-body temperature (optical) was noticed within $\sim$2 days after the explosion, which successively decreased, indicating shock break out inside a dense circumstellar medium (CSM) surrounding the progenitor. Using semianalytical modeling, the ejecta mass and progenitor radius were estimated as 1.2 $M_\odot$ and $\sim$550 $R_\odot$. The archival deep images ($g,r,i and z$ bands) from the Dark Energy Camera Legacy Survey were examined, and a possible progenitor was detected in each band ($\sim$22--22.5 mag) and had a mass range of 14--17 $M_\odot$.

astro-ph.HE↗

arXiv2024

LVCP: LiDAR-Vision Tightly Coupled Collaborative Real-time Relative Positioning

In air-ground collaboration scenarios without GPS and prior maps, the relative positioning of drones and unmanned ground vehicles (UGVs) has always been a challenge. For a drone equipped with monocular camera and an UGV equipped with LiDAR as an external sensor, we propose a robust and real-time relative pose estimation method (LVCP) based on the tight coupling of vision and LiDAR point cloud information, which does not require prior information such as maps or precise initial poses. Given that large-scale point clouds generated by 3D sensors has more accurate spatial geometric information than the feature point cloud generated by image, we utilize LiDAR point clouds to correct the drift in visual-inertial odometry (VIO) when the camera undergoes significant shaking or the IMU has a low signal-to-noise ratio. To achieve this, we propose a novel coarse-to-fine framework for LiDAR-vision collaborative localization. In this framework, we construct point-plane association based on spatial geometric information, and innovatively construct a point-aided Bundle Adjustment (BA) problem as the backend to simultaneously estimate the relative pose of the camera and LiDAR and correct the VIO drift. In this process, we propose a particle swarm optimization (PSO) based sampling algorithm to complete the coarse estimation of the current camera-LiDAR pose. In this process, the initial pose of the camera used for sampling is obtained based on VIO propagation, and the valid feature-plane association number (VFPN) is used to trigger PSO-sampling process. Additionally, we propose a method that combines Structure from Motion (SFM) and multi-level sampling to initialize the algorithm, addressing the challenge of lacking initial values.

cs.RO↗

arXiv2024

Mask-coding-assisted continuous-variable quantum direct communication with orbital angular momentum multiplexing

Quantum secure direct communication (QSDC) is a approach of communication to transmit secret messages base on quantum mechanics. Different from the quantum key distribution, secret messages can be transmitted directly on quantum channel with QSDC. Higher channel capacity and noise suppression capabilities are key to achieving long-distance quantum communication. Here we report a continuous-variable QSDC scheme based on mask-coding and orbital angular momentum, in which the mask-coding is employed to protect the security of the transmitting messages and to suppression the influence of excess noise. And the combination of orbital angular momentum and the information block transmission can effectively improve the secrecy capacity. In the 800 information blocks * 1310 bits length 10-km experiment, the results show that the statistical average of bit error rate has decreased by 36% and reached 0.38% comparing to existing solutions, and the system excess noise is 0.0184 SNU, and the final secrecy capacity is 6.319 * 106 bps. Therefore, this scheme reduces error bits while increasing secrecy capacity, providing a solution for long-distance large-scale quantum communication.

quant-ph↗

arXiv2024

Multiband Simultaneous Photometry of Type II SN 2023ixf with Mephisto and the Twin 50-cm Telescopes

SN 2023ixf, recently reported in the nearby galaxy M101 at a distance of $6.85~{\rm Mpc}$, was one of the closest and brightest core-collapse supernovae (CCSNe) in the last decade. In this work, we present multi-wavelength photometric observation of SN 2023ixf with the Multi-channel Photometric Survey Telescope (Mephisto) in $uvgr$ bands and with the twin 50-cm telescopes in $griz$ bands. We find that the bolometric luminosity reached the maximum value of $3\times10^{43}~{\rm erg~s^{-1}}$ at 3.9 days after the explosion and fully settled onto the radioactive tail at $\sim90$ days. The effective temperature decreased from $3.2\times10^4~{\rm K}$ at the first observation and approached to a constant of $\sim(3000-4000)~{\rm K}$ after the first two months. The evolution of the photospheric radius is consistent with a homologous expansion with a velocity of $8700~{\rm km~s^{-1}}$ in the first two months, and it shrunk subsequently. Based on the radioactive tail, the initial nickel mass is about $M_{\rm Ni}\sim 0.098M_\odot$. The explosion energy and the ejecta mass are estimated to be $E\simeq(1.0-5.7)\times10^{51}~{\rm erg}$ and $M_{\rm ej}\simeq(3.8-16)M_\odot$, respectively. The peak bolometric luminosity is proposed to be contributed by the interaction between the ejecta and the circumstellar medium (CSM). We find a shocked CSM mass of $M_{\rm CSM}\sim0.013M_\odot$, a CSM density of $ρ_{\rm CSM}\sim2.5\times10^{-13}~{\rm g~cm^{-3}}$ and a mass loss rate of the progenitor of $\dot M\sim0.022M_\odot~{\rm yr^{-1}}$.

astro-ph.HE↗

arXiv2024

Revisiting Feature Prediction for Learning Visual Representations from Video

This paper explores feature prediction as a stand-alone objective for unsupervised learning from video and introduces V-JEPA, a collection of vision models trained solely using a feature prediction objective, without the use of pretrained image encoders, text, negative examples, reconstruction, or other sources of supervision. The models are trained on 2 million videos collected from public datasets and are evaluated on downstream image and video tasks. Our results show that learning by predicting video features leads to versatile visual representations that perform well on both motion and appearance-based tasks, without adaption of the model's parameters; e.g., using a frozen backbone. Our largest model, a ViT-H/16 trained only on videos, obtains 81.9% on Kinetics-400, 72.2% on Something-Something-v2, and 77.9% on ImageNet1K.

cs.CV↗

arXiv2024

TransformLoc: Transforming MAVs into Mobile Localization Infrastructures in Heterogeneous Swarms

A heterogeneous micro aerial vehicles (MAV) swarm consists of resource-intensive but expensive advanced MAVs (AMAVs) and resource-limited but cost-effective basic MAVs (BMAVs), offering opportunities in diverse fields. Accurate and real-time localization is crucial for MAV swarms, but current practices lack a low-cost, high-precision, and real-time solution, especially for lightweight BMAVs. We find an opportunity to accomplish the task by transforming AMAVs into mobile localization infrastructures for BMAVs. However, turning this insight into a practical system is non-trivial due to challenges in location estimation with BMAVs' unknown and diverse localization errors and resource allocation of AMAVs given coupled influential factors. This study proposes TransformLoc, a new framework that transforms AMAVs into mobile localization infrastructures, specifically designed for low-cost and resource-constrained BMAVs. We first design an error-aware joint location estimation model to perform intermittent joint location estimation for BMAVs and then design a proximity-driven adaptive grouping-scheduling strategy to allocate resources of AMAVs dynamically. TransformLoc achieves a collaborative, adaptive, and cost-effective localization system suitable for large-scale heterogeneous MAV swarms. We implement TransformLoc on industrial drones and validate its performance. Results show that TransformLoc outperforms baselines including SOTA up to 68\% in localization performance, motivating up to 60\% navigation success rate improvement.

cs.NI↗