Current self-supervised learning algorithms are often modality-specific and require large amounts of computational resources. To address these issues, we increase the training efficiency of data2vec, a learning objective that generalizes across several modalities. We do not encode masked tokens, use a fast convolutional decoder and amortize the effort to build teacher representations. data2vec 2.0 benefits from the rich contextualized target representations introduced in data2vec which enable a fast self-supervised learner. Experiments on ImageNet-1K image classification show that data2vec 2.0 matches the accuracy of Masked Autoencoders in 16.4x lower pre-training time, on Librispeech speech recognition it performs as well as wav2vec 2.0 in 10.6x less time, and on GLUE natural language understanding it matches a retrained RoBERTa model in half the time. Trading some speed for accuracy results in ImageNet-1K top-1 accuracy of 86.8\% with a ViT-L model trained for 150 epochs.
Handwritten Text Recognition (HTR) is more interesting and challenging than printed text due to uneven variations in the handwriting style of the writers, content, and time. HTR becomes more challenging for the Indic languages because of (i) multiple characters combined to form conjuncts which increase the number of characters of respective languages, and (ii) near to 100 unique basic Unicode characters in each Indic script. Recently, many recognition methods based on the encoder-decoder framework have been proposed to handle such problems. They still face many challenges, such as image blur and incomplete characters due to varying writing styles and ink density. We argue that most encoder-decoder methods are based on local visual features without explicit global semantic information. In this work, we enhance the performance of Indic handwritten text recognizers using global semantic information. We use a semantic module in an encoder-decoder framework for extracting global semantic information to recognize the Indic handwritten texts. The semantic information is used in both the encoder for supervision and the decoder for initialization. The semantic information is predicted from the word embedding of a pre-trained language model. Extensive experiments demonstrate that the proposed framework achieves state-of-the-art results on handwritten texts of ten Indic languages.
Accurate airway extraction from computed tomography (CT) images is a critical step for planning navigation bronchoscopy and quantitative assessment of airway-related chronic obstructive pulmonary disease (COPD). The existing methods are challenging to sufficiently segment the airway, especially the high-generation airway, with the constraint of the limited label and cannot meet the clinical use in COPD. We propose a novel two-stage 3D contextual transformer-based U-Net for airway segmentation using CT images. The method consists of two stages, performing initial and refined airway segmentation. The two-stage model shares the same subnetwork with different airway masks as input. Contextual transformer block is performed both in the encoder and decoder path of the subnetwork to finish high-quality airway segmentation effectively. In the first stage, the total airway mask and CT images are provided to the subnetwork, and the intrapulmonary airway mask and corresponding CT scans to the subnetwork in the second stage. Then the predictions of the two-stage method are merged as the final prediction. Extensive experiments were performed on in-house and multiple public datasets. Quantitative and qualitative analysis demonstrate that our proposed method extracted much more branches and lengths of the tree while accomplishing state-of-the-art airway segmentation performance. The code is available at https://github.com/zhaozsq/airway_segmentation.
The sensing process of large-scale LiDAR point clouds inevitably causes large blind spots, i.e. regions not visible to the sensor. We demonstrate how these inherent sampling properties can be effectively utilized for self-supervised representation learning by designing a highly effective pre-training framework that considerably reduces the need for tedious 3D annotations to train state-of-the-art object detectors. Our Masked AutoEncoder for LiDAR point clouds (MAELi) intuitively leverages the sparsity of LiDAR point clouds in both the encoder and decoder during reconstruction. This results in more expressive and useful initialization, which can be directly applied to downstream perception tasks, such as 3D object detection or semantic segmentation for autonomous driving. In a novel reconstruction approach, MAELi distinguishes between empty and occluded space and employs a new masking strategy that targets the LiDAR's inherent spherical projection. Thereby, without any ground truth whatsoever and trained on single frames only, MAELi obtains an understanding of the underlying 3D scene geometry and semantics. To demonstrate the potential of MAELi, we pre-train backbones in an end-to-end manner and show the effectiveness of our unsupervised pre-trained weights on the tasks of 3D object detection and semantic segmentation.
Accurate traffic flow prediction, a hotspot for intelligent transportation research, is the prerequisite for mastering traffic and making travel plans. The speed of traffic flow can be affected by roads condition, weather, holidays, etc. Furthermore, the sensors to catch the information about traffic flow will be interfered with by environmental factors such as illumination, collection time, occlusion, etc. Therefore, the traffic flow in the practical transportation system is complicated, uncertain, and challenging to predict accurately. This paper proposes a deep encoder-decoder prediction framework based on variational Bayesian inference. A Bayesian neural network is constructed by combining variational inference with gated recurrent units (GRU) and used as the deep neural network unit of the encoder-decoder framework to mine the intrinsic dynamics of traffic flow. Then, the variational inference is introduced into the multi-head attention mechanism to avoid noise-induced deterioration of prediction accuracy. The proposed model achieves superior prediction performance on the Guangzhou urban traffic flow dataset over the benchmarks, particularly when the long-term prediction.
Quantifying motion in 3D is important for studying the behavior of humans and other animals, but manual pose annotations are expensive and time-consuming to obtain. Self-supervised keypoint discovery is a promising strategy for estimating 3D poses without annotations. However, current keypoint discovery approaches commonly process single 2D views and do not operate in the 3D space. We propose a new method to perform self-supervised keypoint discovery in 3D from multi-view videos of behaving agents, without any keypoint or bounding box supervision in 2D or 3D. Our method, BKinD-3D, uses an encoder-decoder architecture with a 3D volumetric heatmap, trained to reconstruct spatiotemporal differences across multiple views, in addition to joint length constraints on a learned 3D skeleton of the subject. In this way, we discover keypoints without requiring manual supervision in videos of humans and rats, demonstrating the potential of 3D keypoint discovery for studying behavior.
Deep Neural Networks (DNNs) suffer from domain shift when the test dataset follows a distribution different from the training dataset. Domain generalization aims to tackle this issue by learning a model that can generalize to unseen domains. In this paper, we propose a new approach that aims to explicitly remove domain-specific features for domain generalization. Following this approach, we propose a novel framework called Learning and Removing Domain-specific features for Generalization (LRDG) that learns a domain-invariant model by tactically removing domain-specific features from the input images. Specifically, we design a classifier to effectively learn the domain-specific features for each source domain, respectively. We then develop an encoder-decoder network to map each input image into a new image space where the learned domain-specific features are removed. With the images output by the encoder-decoder network, another classifier is designed to learn the domain-invariant features to conduct image classification. Extensive experiments demonstrate that our framework achieves superior performance compared with state-of-the-art methods.
Traditional multi-task learning architectures train a single model across multiple tasks through a shared encoder followed by task-specific decoders. Learning these models often requires specialized training algorithms that address task-conflict in the shared parameter updates, which otherwise can lead to negative transfer. A new type of multi-task learning within NLP homogenizes multi-task architectures as a shared encoder and language model decoder, which does surprisingly well across a range of diverse tasks. Does this new architecture suffer from task-conflicts that require specialized training algorithms? We study how certain factors in the shift towards text-to-text models affects multi-task conflict and negative transfer, finding that both directional conflict and transfer are surprisingly constant across architectures.
This paper presents a Long Range (LoRa) physical-layer data aggregation system (LoRaPDA) that aggregates data (e.g., sum, average, min, max) directly in the physical layer. In particular, after coordinating a few nodes to transmit their data simultaneously, the gateway leverages a new multi-packet reception (MPR) approach to compute aggregate data from the phase-asynchronous superimposed signal. Different from the analog approach which requires additional power synchronization and phase synchronization, our MRP-based digital approach is compatible with commercial LoRa nodes and is more reliable. Different from traditional MPR approaches that are designed for the collision decoding scenario, our new MPR approach allows simultaneous transmissions with small packet arrival time offsets, and addresses a new co-located peak problem through the following components: 1) an improved channel and offset estimation algorithm that enables accurate phase tracking in each symbol, 2) a new symbol demodulation algorithm that finds the maximum likelihood sequence of nodes' data, and 3) a soft-decision packet decoding algorithm that utilizes the likelihoods of several sequences to improve decoding performance. Trace-driven simulation results show that the symbol demodulation algorithm outperforms the state-of-the-art MPR decoder by 5.3$\times$ in terms of physical-layer throughput, and the soft decoder is more robust to unavoidable adverse phase misalignment and estimation error in practice. Moreover, LoRaPDA outperforms the state-of-the-art MPR scheme by at least 2.1$\times$ for all SNRs in terms of network throughput, demonstrating quick and reliable data aggregation.
Recently, the success of pre-training in text domain has been fully extended to vision, audio, and cross-modal scenarios. The proposed pre-training models of different modalities are showing a rising trend of homogeneity in their model structures, which brings the opportunity to implement different pre-training models within a uniform framework. In this paper, we present TencentPretrain, a toolkit supporting pre-training models of different modalities. The core feature of TencentPretrain is the modular design. The toolkit uniformly divides pre-training models into 5 components: embedding, encoder, target embedding, decoder, and target. As almost all of common modules are provided in each component, users can choose the desired modules from different components to build a complete pre-training model. The modular design enables users to efficiently reproduce existing pre-training models or build brand-new one. We test the toolkit on text, vision, and audio benchmarks and show that it can match the performance of the original implementations.
Current approaches to empathetic response generation typically encode the entire dialogue history directly and put the output into a decoder to generate friendly feedback. These methods focus on modelling contextual information but neglect capturing the direct intention of the speaker. We argue that the last utterance in the dialogue empirically conveys the intention of the speaker. Consequently, we propose a novel model named InferEM for empathetic response generation. We separately encode the last utterance and fuse it with the entire dialogue through the multi-head attention based intention fusion module to capture the speaker's intention. Besides, we utilize previous utterances to predict the last utterance, which simulates human's psychology to guess what the interlocutor may speak in advance. To balance the optimizing rates of the utterance prediction and response generation, a multi-task learning strategy is designed for InferEM. Experimental results demonstrate the plausibility and validity of InferEM in improving empathetic expression.
Laminated composite materials are widely used in most fields of engineering. Wave propagation analysis plays an essential role in understanding the short-duration transient response of composite structures. The forward physics-based models are utilized to map from elastic properties space to wave propagation behavior in a laminated composite material. Due to the high-frequency, multi-modal, and dispersive nature of the guided waves, the physics-based simulations are computationally demanding. It makes property prediction, generation, and material design problems more challenging. In this work, a forward physics-based simulator such as the stiffness matrix method is utilized to collect group velocities of guided waves for a set of composite materials. A variational autoencoder (VAE)-based deep generative model is proposed for the generation of new and realistic polar group velocity representations. It is observed that the deep generator is able to reconstruct unseen representations with very low mean square reconstruction error. Global Monte Carlo and directional equally-spaced samplers are used to sample the continuous, complete and organized low-dimensional latent space of VAE. The sampled point is fed into the trained decoder to generate new polar representations. The network has shown exceptional generation capabilities. It is also seen that the latent space forms a conceptual space where different directions and regions show inherent patterns related to the generated representations and their corresponding material properties.
We present the Group Propagation Vision Transformer (GPViT): a novel nonhierarchical (i.e. non-pyramidal) transformer model designed for general visual recognition with high-resolution features. High-resolution features (or tokens) are a natural fit for tasks that involve perceiving fine-grained details such as detection and segmentation, but exchanging global information between these features is expensive in memory and computation because of the way self-attention scales. We provide a highly efficient alternative Group Propagation Block (GP Block) to exchange global information. In each GP Block, features are first grouped together by a fixed number of learnable group tokens; we then perform Group Propagation where global information is exchanged between the grouped features; finally, global information in the updated grouped features is returned back to the image features through a transformer decoder. We evaluate GPViT on a variety of visual recognition tasks including image classification, semantic segmentation, object detection, and instance segmentation. Our method achieves significant performance gains over previous works across all tasks, especially on tasks that require highresolution outputs, for example, our GPViT-L3 outperforms Swin Transformer-B by 2.0 mIoU on ADE20K semantic segmentation with only half as many parameters. Project page: chenhongyiyang.com/projects/GPViT/GPViT
Compact and accurate representations of 3D shapes are central to many perception and robotics tasks. State-of-the-art learning-based methods can reconstruct single objects but scale poorly to large datasets. We present a novel recursive implicit representation to efficiently and accurately encode large datasets of complex 3D shapes by recursively traversing an implicit octree in latent space. Our implicit Recursive Octree Auto-Decoder (ROAD) learns a hierarchically structured latent space enabling state-of-the-art reconstruction results at a compression ratio above 99%. We also propose an efficient curriculum learning scheme that naturally exploits the coarse-to-fine properties of the underlying octree spatial representation. We explore the scaling law relating latent space dimension, dataset size, and reconstruction accuracy, showing that increasing the latent space dimension is enough to scale to large shape datasets. Finally, we show that our learned latent space encodes a coarse-to-fine hierarchical structure yielding reusable latents across different levels of details, and we provide qualitative evidence of generalization to novel shapes outside the training set.
Pre-training by numerous image data has become de-facto for robust 2D representations. In contrast, due to the expensive data acquisition and annotation, a paucity of large-scale 3D datasets severely hinders the learning for high-quality 3D features. In this paper, we propose an alternative to obtain superior 3D representations from 2D pre-trained models via Image-to-Point Masked Autoencoders, named as I2P-MAE. By self-supervised pre-training, we leverage the well learned 2D knowledge to guide 3D masked autoencoding, which reconstructs the masked point tokens with an encoder-decoder architecture. Specifically, we first utilize off-the-shelf 2D models to extract the multi-view visual features of the input point cloud, and then conduct two types of image-to-point learning schemes on top. For one, we introduce a 2D-guided masking strategy that maintains semantically important point tokens to be visible for the encoder. Compared to random masking, the network can better concentrate on significant 3D structures and recover the masked tokens from key spatial cues. For another, we enforce these visible tokens to reconstruct the corresponding multi-view 2D features after the decoder. This enables the network to effectively inherit high-level 2D semantics learned from rich image data for discriminative 3D modeling. Aided by our image-to-point pre-training, the frozen I2P-MAE, without any fine-tuning, achieves 93.4% accuracy for linear SVM on ModelNet40, competitive to the fully trained results of existing methods. By further fine-tuning on on ScanObjectNN's hardest split, I2P-MAE attains the state-of-the-art 90.11% accuracy, +3.68% to the second-best, demonstrating superior transferable capacity. Code will be available at https://github.com/ZrrSkywalker/I2P-MAE.
How to identify and segment camouflaged objects from the background is challenging. Inspired by the multi-head self-attention in Transformers, we present a simple masked separable attention (MSA) for camouflaged object detection. We first separate the multi-head self-attention into three parts, which are responsible for distinguishing the camouflaged objects from the background using different mask strategies. Furthermore, we propose to capture high-resolution semantic representations progressively based on a simple top-down decoder with the proposed MSA to attain precise segmentation results. These structures plus a backbone encoder form a new model, dubbed CamoFormer. Extensive experiments show that CamoFormer surpasses all existing state-of-the-art methods on three widely-used camouflaged object detection benchmarks. There are on average around 5% relative improvements over previous methods in terms of S-measure and weighted F-measure.
As the complexity of modern manufacturing technologies increases, traditional trial-and-error design, which requires iterative and expensive simulations, becomes unreliable and time-consuming. This difficulty is especially significant for the design of hot-stamped safety-critical components, such as ultra-high-strength-steel (UHSS) B-pillars. To reduce design costs and ensure manufacturability, scalar-based Artificial-Intelligence-empowered surrogate modelling (SAISM) has been investigated and implemented, which can allow real-time manufacturability-constrained structural design optimisation. However, SAISM suffers from low accuracy and generalisability, and usually requires a high volume of training samples. To solve this problem, an image-based Artificial-intelligence-empowered surrogate modelling (IAISM) approach is developed in this research, in combination with an auto-decoder-based blank shape generator. The IAISM, which is based on a Mask-Res-SE-U-Net architecture, is trained to predict the full thinning field of the as-formed component given an arbitrary blank shape. Excellent prediction performance of IAISM is achieved with only 256 training samples, which indicates the small-data learning nature of engineering AI tasks using structured data representations. The trained auto-decoder, trained Mask-Res-SE-U-Net, and Adam optimiser are integrated to conduct blank optimisation by modifying the latent vector. The optimiser can rapidly find blank shapes that satisfy manufacturability criteria. As a high-accuracy and generalisable surrogate modelling and optimisation tool, the proposed pipeline is promising to be integrated into a full-chain digital twin to conduct real-time, multi-objective design optimisation.
Although understanding and characterizing causal effects have become essential in observational studies, it is challenging when the confounders are high-dimensional. In this article, we develop a general framework $\textit{CausalEGM}$ for estimating causal effects by encoding generative modeling, which can be applied in both binary and continuous treatment settings. Under the potential outcome framework with unconfoundedness, we establish a bidirectional transformation between the high-dimensional confounders space and a low-dimensional latent space where the density is known (e.g., multivariate normal distribution). Through this, CausalEGM simultaneously decouples the dependencies of confounders on both treatment and outcome and maps the confounders to the low-dimensional latent space. By conditioning on the low-dimensional latent features, CausalEGM can estimate the causal effect for each individual or the average causal effect within a population. Our theoretical analysis shows that the excess risk for CausalEGM can be bounded through empirical process theory. Under an assumption on encoder-decoder networks, the consistency of the estimate can be guaranteed. In a series of experiments, CausalEGM demonstrates superior performance over existing methods for both binary and continuous treatments. Specifically, we find CausalEGM to be substantially more powerful than competing methods in the presence of large sample sizes and high dimensional confounders. The software of CausalEGM is freely available at https://github.com/SUwonglab/CausalEGM.
In this paper, we propose a control synthesis method for signal temporal logic (STL) specifications with neural networks (NNs). Most of the previous works consider training a controller for only a given STL specification. These approaches, however, require retraining the NN controller if a new specification arises and needs to be satisfied, which results in large consumption of memory and inefficient training. To tackle this problem, we propose to construct NN controllers by introducing encoder-decoder structured NNs with an attention mechanism. The encoder takes an STL formula as input and encodes it into an appropriate vector, and the decoder outputs control signals that will meet the given specification. As the encoder, we consider three NN structures: sequential, tree-structured, and graph-structured NNs. All the model parameters are trained in an end-to-end manner to maximize the expected robustness that is known to be a quantitative semantics of STL formulae. We compare the control performances attained by the above NN structures through a numerical experiment of the path planning problem, showing the efficacy of the proposed approach.
An insight into the architecture of the Encoder-Decoder Network with Guided Transmission Map (EDN-GTM), a novel and effective single image dehazing scheme, is presented in this paper. The EDN-GTM takes a conventional RGB hazy image in conjunction with the corresponding transmission map estimated by the dark channel prior (DCP) approach as inputs of the network. The EDN-GTM adopts an enhanced structure of U-Net developed for dehazing tasks and the resulting EDN-GDM has shown state-of-the-art performances on benchmark dehazing datasets in terms of PSNR and SSIM metrics. In order to give an in-depth understanding of the well-designed architecture which largely contributes to the success of the EDN-GTM, extensive experiments and analysis from selecting the core structure of the scheme to investigating advanced network designs are presented in this paper.