Search

Fully quantized training (FQT) accelerates the training of deep neural networks by quantizing the activations, weights, and gradients into lower precision. To explore the ultimate limit of FQT (the lowest achievable precision), we make a first attempt to 1-bit FQT. We provide a theoretical analysis of FQT based on Adam and SGD, revealing that the gradient variance influences the convergence of FQT. Building on these theoretical results, we introduce an Activation Gradient Pruning (AGP) strategy. The strategy leverages the heterogeneity of gradients by pruning less informative gradients and enhancing the numerical precision of remaining gradients to mitigate gradient variance. Additionally, we propose Sample Channel joint Quantization (SCQ), which utilizes different quantization strategies in the computation of weight gradients and activation gradients to ensure that the method is friendly to low-bitwidth hardware. Finally, we present a framework to deploy our algorithm. For fine-tuning VGGNet-16 and ResNet-18 on multiple datasets, our algorithm achieves an average accuracy improvement of approximately 6%, compared to per-sample quantization. Moreover, our training speedup can reach a maximum of 5.13x compared to full precision training.

Talking head synthesis is a practical technique with wide applications. Current Neural Radiance Field (NeRF) based approaches have shown their superiority on driving one-shot talking heads with videos or signals regressed from audio. However, most of them failed to take the audio as driven information directly, unable to enjoy the flexibility and availability of speech. Since mapping audio signals to face deformation is non-trivial, we design a Single-Shot Speech-Driven Neural Radiance Field (S^3D-NeRF) method in this paper to tackle the following three difficulties: learning a representative appearance feature for each identity, modeling motion of different face regions with audio, and keeping the temporal consistency of the lip area. To this end, we introduce a Hierarchical Facial Appearance Encoder to learn multi-scale representations for catching the appearance of different speakers, and elaborate a Cross-modal Facial Deformation Field to perform speech animation according to the relationship between the audio signal and different face regions. Moreover, to enhance the temporal consistency of the important lip area, we introduce a lip-sync discriminator to penalize the out-of-sync audio-visual sequences. Extensive experiments have shown that our S^3D-NeRF surpasses previous arts on both video fidelity and audio-lip synchronization., Comment: ECCV 2024

While Large Language Models (LLMs) have achieved remarkable success in various fields, the efficiency of training and inference remains a major challenge. To address this issue, we propose SUBLLM, short for Subsampling-Upsampling-Bypass Large Language Model, an innovative architecture that extends the core decoder-only framework by incorporating subsampling, upsampling, and bypass modules. The subsampling modules are responsible for shortening the sequence, while the upsampling modules restore the sequence length, and the bypass modules enhance convergence. In comparison to LLaMA, the proposed SUBLLM exhibits significant enhancements in both training and inference speeds as well as memory usage, while maintaining competitive few-shot performance. During training, SUBLLM increases speeds by 26% and cuts memory by 10GB per GPU. In inference, it boosts speeds by up to 37% and reduces memory by 1GB per GPU. The training and inference speeds can be enhanced by 34% and 52% respectively when the context window is expanded to 8192. Our code is available at https://github.com/XiaoMi/subllm., Comment: 10 pages, 5 figures, accepted by ECAI 2024

We conduct a comprehensive analysis of the accreting X-ray pulsar, Hercules X-1, utilizing data from IXPE and NuSTAR. IXPE performed five observations of Her X-1, consisting of three in the Main-on state and two in the Short-on state. Our time-resolved analysis uncovers the linear correlations between the flux and polarization degree as well as the pulse fraction and polarization degree. Geometry parameters are rigorously constrained by fitting the phase-resolved modulations of Cyclotron Resonance Scattering Feature and polarization angle with a simple dipole model and Rotating Vector Model respectively, yielding roughly consistent results. The changes of $\chi_{\rm p}$ (the position angle of the pulsar's spin axis on the plane of the sky) between different Main-on observations suggest the possible forced precession of the neutron star crust. Furthermore, a linear association between the energy of Cyclotron Resonance Scattering Feature and polarization angle implies the prevalence of a dominant dipole magnetic field, and their phase-resolved modulations likely arise from viewing angle effects., Comment: Accepted for MNRAS

Training large transformers is slow, but recent innovations on GPU architecture give us an advantage. NVIDIA Ampere GPUs can execute a fine-grained 2:4 sparse matrix multiplication twice as fast as its dense equivalent. In the light of this property, we comprehensively investigate the feasibility of accelerating feed-forward networks (FFNs) of transformers in pre-training. First, we define a ``flip rate'' to monitor the stability of a 2:4 training process. Utilizing this metric, we propose three techniques to preserve accuracy: to modify the sparse-refined straight-through estimator by applying the masked decay term on gradients, to determine a feasible decay factor in warm-up stage, and to enhance the model's quality by a dense fine-tuning procedure near the end of pre-training. Besides, we devise two techniques to practically accelerate training: to calculate transposable 2:4 masks by convolution, and to accelerate gated activation functions by reducing GPU L2 cache miss. Experiments show that our 2:4 sparse training algorithm achieves similar convergence to dense training algorithms on several transformer pre-training tasks, while actual acceleration can be observed on different shapes of transformer block apparently. Our toolkit is available at https://github.com/huyz2023/2by4-pretrain.

Pretraining transformers are generally time-consuming. Fully quantized training (FQT) is a promising approach to speed up pretraining. However, most FQT methods adopt a quantize-compute-dequantize procedure, which often leads to suboptimal speedup and significant performance degradation when used in transformers due to the high memory access overheads and low-precision computations. In this work, we propose Jetfire, an efficient and accurate INT8 training method specific to transformers. Our method features an INT8 data flow to optimize memory access and a per-block quantization method to maintain the accuracy of pretrained transformers. Extensive experiments demonstrate that our INT8 FQT method achieves comparable accuracy to the FP16 training baseline and outperforms the existing INT8 training works for transformers. Moreover, for a standard transformer block, our method offers an end-to-end training speedup of 1.42x and a 1.49x memory reduction compared to the FP16 baseline. Our code is open sourced at https://github.com/thu-ml/Jetfire-INT8Training., Comment: 15 pages, 8 figures, 11 tables

The Chasing All Transients Constellation Hunters (CATCH) space mission is an intelligent constellation consisting of 126 micro-satellites in three types (A, B, and C), designed for X-ray observation with the objective of studying the dynamic universe. Currently, we are actively developing the first Pathfinder (CATCH-1) for the CATCH mission, specifically for type-A satellites. CATCH-1 is equipped with Micro Pore Optics (MPO) and a 4-pixel Silicon Drift Detector (SDD) array. To assess its scientific performance, including the effective area of the optical system, on-orbit background, and telescope sensitivity, we employ the Monte Carlo software Geant4 for simulation in this study. The MPO optics exhibit an effective area of $41$ cm$^2$ at the focal spot for 1 keV X-rays, while the entire telescope system achieves an effective area of $29$ cm$^2$ at 1 keV when taking into account the SDD detector's detection efficiency. The primary contribution to the background is found to be from the Cosmic X-ray Background. Assuming a 625 km orbit with an inclination of $29^\circ$, the total background for CATCH-1 is estimated to be $8.13\times10^{-2}$ counts s$^{-1}$ in the energy range of 0.5--4 keV. Based on the background within the central detector and assuming a Crab-like source spectrum, the estimated ideal sensitivity could achieve $1.9\times10^{-12}$ erg cm$^{-2}$ s$^{-1}$ for an exposure of 10$^4$ s in the energy band of 0.5--4 keV. Furthermore, after simulating the background caused by low-energy charged particles near the geomagnetic equator, we have determined that there is no need to install a magnetic deflector.

Audio-driven talking head synthesis is a promising topic with wide applications in digital human, film making and virtual reality. Recent NeRF-based approaches have shown superiority in quality and fidelity compared to previous studies. However, when it comes to few-shot talking head generation, a practical scenario where only few seconds of talking video is available for one identity, two limitations emerge: 1) they either have no base model, which serves as a facial prior for fast convergence, or ignore the importance of audio when building the prior; 2) most of them overlook the degree of correlation between different face regions and audio, e.g., mouth is audio related, while ear is audio independent. In this paper, we present Audio Enhanced Neural Radiance Field (AE-NeRF) to tackle the above issues, which can generate realistic portraits of a new speaker with fewshot dataset. Specifically, we introduce an Audio Aware Aggregation module into the feature fusion stage of the reference scheme, where the weight is determined by the similarity of audio between reference and target image. Then, an Audio-Aligned Face Generation strategy is proposed to model the audio related and audio independent regions respectively, with a dual-NeRF framework. Extensive experiments have shown AE-NeRF surpasses the state-of-the-art on image fidelity, audio-lip synchronization, and generalization ability, even in limited training set or training iterations., Comment: Accepted by AAAI 2024

Video synthesis has recently made remarkable strides benefiting from the rapid development of diffusion models. However, it still encounters challenges in terms of semantic accuracy, clarity and spatio-temporal continuity. They primarily arise from the scarcity of well-aligned text-video data and the complex inherent structure of videos, making it difficult for the model to simultaneously ensure semantic and qualitative excellence. In this report, we propose a cascaded I2VGen-XL approach that enhances model performance by decoupling these two factors and ensures the alignment of the input data by utilizing static images as a form of crucial guidance. I2VGen-XL consists of two stages: i) the base stage guarantees coherent semantics and preserves content from input images by using two hierarchical encoders, and ii) the refinement stage enhances the video's details by incorporating an additional brief text and improves the resolution to 1280$\times$720. To improve the diversity, we collect around 35 million single-shot text-video pairs and 6 billion text-image pairs to optimize the model. By this means, I2VGen-XL can simultaneously enhance the semantic accuracy, continuity of details and clarity of generated videos. Through extensive experiments, we have investigated the underlying principles of I2VGen-XL and compared it with current top methods, which can demonstrate its effectiveness on diverse data. The source code and models will be publicly available at \url{https://i2vgen-xl.github.io}., Comment: Project page: https://i2vgen-xl.github.io

Text-guided image editing faces significant challenges to training and inference flexibility. Much literature collects large amounts of annotated image-text pairs to train text-conditioned generative models from scratch, which is expensive and not efficient. After that, some approaches that leverage pre-trained vision-language models are put forward to avoid data collection, but they are also limited by either per text-prompt optimization or inference-time hyper-parameters tuning. To address these issues, we investigate and identify a specific space, referred to as CLIP DeltaSpace, where the CLIP visual feature difference of two images is semantically aligned with the CLIP textual feature difference of their corresponding text descriptions. Based on DeltaSpace, we propose a novel framework called DeltaEdit, which maps the CLIP visual feature differences to the latent space directions of a generative model during the training phase, and predicts the latent space directions from the CLIP textual feature differences during the inference phase. And this design endows DeltaEdit with two advantages: (1) text-free training; (2) generalization to various text prompts for zero-shot inference. Extensive experiments validate the effectiveness and versatility of DeltaEdit with different generative models, including both the GAN model and the diffusion model, in achieving flexible text-guided image editing. Code is available at https://github.com/Yueming6568/DeltaEdit., Comment: 17 pages. arXiv admin note: text overlap with arXiv:2303.06285

We present the results of a detailed timing and spectral analysis of the quasi-regular modulation (QRM) phenomenon in the black hole X-ray binary 4U 1630--47 during its 1998 outburst observed by Rossi X-ray Timing Explore (RXTE). We find that the $\sim$ 50-110 mHz QRM is flux dependent, and the QRM is detected with simultaneous low frequency quasi-periodic oscillations (LFQPOs). According to the behavior of the power density spectrum, we divide the observations into four groups. In the first group, namely behavior A, LFQPOs are detected, but no mHz QRM. The second group, namely behavior B, a QRM with frequency above $\sim$ 88 mHz is detected and the $\sim$ 5 Hz and $\sim$ 7 Hz LFQPOs are almost overlapping. In the third group, namely behavior C, the QRM frequency below $\sim$ 88 mHz is detected and the LFQPOs are significantly separated. In the forth group, namely behavior D, neither QRM nor LFQPOs are detected. We study the energy-dependence of the fractional rms, centroid frequency, and phase-lag of QRM and LFQPOs for behavior B and C. We then study the evolution of QRM and find that the frequency of QRM increases with hardness, while its rms decreases with hardness. We also analyze the spectra of each observation, and find that the QRM rms of behavior B has a positive correlation with $\rm F_{\rm powerlaw}$ / $\rm F_{\rm total}$. Finally, we give our understanding for this mHz QRM phenomena., Comment: 14pages, 15 figures

Efficiently generating a freestyle 3D portrait with high quality and 3D-consistency is a promising yet challenging task. The portrait styles generated by most existing methods are usually restricted by their 3D generators, which are learned in specific facial datasets, such as FFHQ. To get the diverse 3D portraits, one can build a large-scale multi-style database to retrain a 3D-aware generator, or use a off-the-shelf tool to do the style translation. However, the former is time-consuming due to data collection and training process, the latter may destroy the multi-view consistency. To tackle this problem, we propose a novel text-driven 3D-aware portrait synthesis framework that can generate out-of-distribution portrait styles. Specifically, for a given portrait style prompt, we first composite two generative priors, a 3D-aware GAN generator and a text-guided image editor, to quickly construct a few-shot stylized portrait set. Then we map the special style domain of this set to our proposed 3D latent feature generator and obtain a 3D representation containing the given style information. Finally we use a pre-trained 3D renderer to generate view-consistent stylized portraits from the 3D representation. Extensive experimental results show that our method is capable of synthesizing high-quality 3D portraits with specified styles in a few minutes, outperforming the state-of-the-art., Comment: project website: https://tianxiangma.github.io/FF3D

Open-domain long-term memory conversation can establish long-term intimacy with humans, and the key is the ability to understand and memorize long-term dialogue history information. Existing works integrate multiple models for modelling through a pipeline, which ignores the coupling between different stages. In this paper, we propose a Unified framework for Long-term Memory Conversations (UniMC), which increases the connection between different stages by learning relevance representation. Specifically, we decompose the main task into three subtasks based on probability graphs: 1) conversation summarization, 2) memory retrieval, 3) memory-augmented generation. Each subtask involves learning a representation for calculating the relevance between the query and memory, which is modelled by inserting a special token at the beginning of the decoder input. The relevance representation learning strengthens the connection across subtasks through parameter sharing and joint training. Extensive experimental results show that the proposed method consistently improves over strong baselines and yields better dialogue consistency and engagingness.

The Chasing All Transients Constellation Hunters (CATCH) space mission plans to launch three types of micro-satellites (A, B, and C). The type-B CATCH satellites are dedicated to locating transients and detecting their time-dependent energy spectra. A type-B satellite is equipped with lightweight Wolter-I X-ray optics and an array of position-sensitive multi-pixel Silicon Drift Detectors. To optimize the scientific payloads for operating properly in orbit and performing the observations with high sensitivities, this work performs an in-orbit background simulation of a type-B CATCH satellite using the Geant4 toolkit. It shows that the persistent background is dominated by the cosmic X-ray diffuse background and the cosmic-ray protons. The dynamic background is also estimated considering trapped charged particles in the radiation belts and low-energy charged particles near the geomagnetic equator, which is dominated by the incident electrons outside the aperture. The simulated persistent background within the focal spot is used to estimate the observation sensitivity, i.e. 4.22$\times$10$^{-13}$ erg cm$^{-2}$ s$^{-1}$ with an exposure of 10$^{4}$ s and a Crab-like source spectrum, which can be utilized further to optimize the shielding design. The simulated in-orbit background also suggests that the magnetic diverter just underneath the optics may be unnecessary in this kind of micro-satellites, because the dynamic background induced by charged particles outside the aperture is around 3 orders of magnitude larger than that inside the aperture., Comment: 24 pages, 13 figures, 7 tables, accepted for publication in Experimental Astronomy

One major issue in learning-based model predictive control (MPC) for autonomous driving is the contradiction between the system model's prediction accuracy and computation efficiency. The more situations a system model covers, the more complex it is, along with highly nonlinear and nonconvex properties. These issues make the optimization too complicated to solve and render real-time control impractical.To address these issues, we propose a hierarchical learning residual model which leverages random forests and linear regression.The learned model consists of two levels. The low level uses linear regression to fit the residues, and the high level uses random forests to switch different linear models. Meanwhile, we adopt the linear dynamic bicycle model with error states as the nominal model.The switched linear regression model is added to the nominal model to form the system model. It reformulates the learning-based MPC as a quadratic program (QP) problem and optimization solvers can effectively solve it. Experimental path tracking results show that the driving vehicle's prediction accuracy and tracking accuracy are significantly improved compared with the nominal MPC.Compared with the state-of-the-art Gaussian process-based nonlinear model predictive control (GP-NMPC), our method gets better performance on tracking accuracy while maintaining a lower computation consumption., Comment: 8 pages, 8 figures

This work presents RiDDLE, short for Reversible and Diversified De-identification with Latent Encryptor, to protect the identity information of people from being misused. Built upon a pre-learned StyleGAN2 generator, RiDDLE manages to encrypt and decrypt the facial identity within the latent space. The design of RiDDLE has three appealing properties. First, the encryption process is cipher-guided and hence allows diverse anonymization using different passwords. Second, the true identity can only be decrypted with the correct password, otherwise the system will produce another de-identified face to maintain the privacy. Third, both encryption and decryption share an efficient implementation, benefiting from a carefully tailored lightweight encryptor. Comparisons with existing alternatives confirm that our approach accomplishes the de-identification task with better quality, higher diversity, and stronger reversibility. We further demonstrate the effectiveness of RiDDLE in anonymizing videos. Code and models will be made publicly available., Comment: Accepted by CVPR 2023

Vision Transformer (ViT) suffers from data scarcity in semi-supervised learning (SSL). To alleviate this issue, inspired by masked autoencoder (MAE), which is a data-efficient self-supervised learner, we propose Semi-MAE, a pure ViT-based SSL framework consisting of a parallel MAE branch to assist the visual representation learning and make the pseudo labels more accurate. The MAE branch is designed as an asymmetric architecture consisting of a lightweight decoder and a shared-weights encoder. We feed the weakly-augmented unlabeled data with a high masking ratio to the MAE branch and reconstruct the missing pixels. Semi-MAE achieves 75.9% top-1 accuracy on ImageNet with 10% labels, surpassing prior state-of-the-art in semi-supervised image classification. In addition, extensive experiments demonstrate that Semi-MAE can be readily used for other ViT models and masked image modeling methods.

In time-domain astronomy, a substantial number of transients will be discovered by multi-wavelength and multi-messenger observatories, posing a great challenge for follow-up capabilities. We have thus proposed an intelligent X-ray constellation, the Chasing All Transients Constellation Hunters (CATCH) space mission. Consisting of 126 micro-satellites in three types, CATCH will have the capability to perform follow-up observations for a large number of different types of transients simultaneously. Each satellite in the constellation will carry lightweight X-ray optics and use a deployable mast to increase the focal length. The combination of different optics and detector systems enables different types of satellites to have multiform observation capabilities, including timing, spectroscopy, imaging, and polarization. Controlled by the intelligent system, different satellites can cooperate to perform uninterrupted monitoring, all-sky follow-up observations, and scanning observations with a flexible field of view (FOV) and multi-dimensional observations. Therefore, CATCH will be a powerful mission to study the dynamic universe. Here, we present the current design of the spacecraft, optics, detector system, constellation configuration and observing modes, as well as the development plan., Comment: Accepted for publication in Experimental Astronomy. 43 pages, 15 figures, 4 tables

By performing P doping on the Si sites in the topological semimetal Mo$_5$Si$_3$, we discover strong-coupling superconductivity in Mo$_5$Si$_{3-x}$P$_x$ (0.5 $\le$ $x$ $\le$ 2.0). Mo$_5$Si$_3$ crystallizes in the W$_5$Si$_3$-type structure with space group of $I4/mcm$ (No. 140), and is not a superconductor itself. Upon P doping, the lattice parameter $a$ decreases while $c$ increases monotonously. Bulk superconductivity is revealed in Mo$_5$Si$_{3-x}$P$_x$ (0.5 $\le$ $x$ $\le$ 2.0) from resistivity, magnetization, and heat capacity measurements. $T_c$ in Mo$_5$Si$_{1.5}$P$_{1.5}$ reaches as high as 10.8 K, setting a new record among the W$_5$Si$_3$-type superconductors. The upper and lower critical fields for Mo$_5$Si$_{1.5}$P$_{1.5}$ are 14.56 T and 105 mT, respectively. Moreover, Mo$_5$Si$_{1.5}$P$_{1.5}$ is found to be a fully gapped superconductor with strong electron-phonon coupling. First-principles calculations suggest that the enhancement of electron-phonon coupling is possibly due to the shift of the Fermi level, which is induced by electron doping. The calculations also reveal the nontrivial band topology in Mo$_5$Si$_3$. The $T_c$ and upper critical field in Mo$_5$Si$_{3-x}$P$_x$ are fairly high among pseudobinary compounds. Both of them are higher than those in NbTi, making future applications promising. Our results suggest that the W$_5$Si$_3$-type compounds are ideal platforms to search for new superconductors. By examinations of their band topologies, more candidates for topological superconductors can be expected in this structural family., Comment: 15 pages, 5 figures. Supplementary Information availabe at the corresponding DOI

Continual relation extraction (CRE) aims to continuously train a model on data with new relations while avoiding forgetting old ones. Some previous work has proved that storing a few typical samples of old relations and replaying them when learning new relations can effectively avoid forgetting. However, these memory-based methods tend to overfit the memory samples and perform poorly on imbalanced datasets. To solve these challenges, a consistent representation learning method is proposed, which maintains the stability of the relation embedding by adopting contrastive learning and knowledge distillation when replaying memory. Specifically, supervised contrastive learning based on a memory bank is first used to train each new task so that the model can effectively learn the relation representation. Then, contrastive replay is conducted of the samples in memory and makes the model retain the knowledge of historical relations through memory knowledge distillation to prevent the catastrophic forgetting of the old task. The proposed method can better learn consistent representations to alleviate forgetting effectively. Extensive experiments on FewRel and TACRED datasets show that our method significantly outperforms state-of-the-art baselines and yield strong robustness on the imbalanced dataset., Comment: Accepted to Findings of ACL 2022