arXiv subjects
Publications and source records attributed to Yu Sun.
Test-time training adapts to a new test distribution on the fly by optimizing a model for each test input using self-supervision. In this paper, we use masked autoencoders for this one-sample learning problem. Empirically, our simple method improves generalization on many visual benchmarks for distribution shifts. Theoretically, we characterize this improvement in terms of the bias-variance trade-off.
Intensity diffraction tomography (IDT) refers to a class of optical microscopy techniques for imaging the 3D refractive index (RI) distribution of a sample from a set of 2D intensity-only measurements. The reconstruction of artifact-free RI maps is a fundamental challenge in IDT due to the loss of phase information and the missing cone problem. Neural fields (NF) has recently emerged as a new deep learning (DL) approach for learning continuous representations of physical fields. NF uses a coordinate-based neural network to represent the field by mapping the spatial coordinates to the corresponding physical quantities, in our case the complex-valued refractive index values. We present DeCAF as the first NF-based IDT method that can learn a high-quality continuous representation of a RI volume from its intensity-only and limited-angle measurements. The representation in DeCAF is learned directly from the measurements of the test sample by using the IDT forward model, without any ground-truth RI maps. We qualitatively and quantitatively evaluate DeCAF on the simulated and experimental biological samples. Our results show that DeCAF can generate high-contrast and artifact-free RI maps and lead to up to 2.1 times reduction in MSE over existing methods.
Flexible task planning continues to pose a difficult challenge for robots, where a robot is unable to creatively adapt their task plans to new or unseen problems, which is mainly due to the limited knowledge it has about its actions and world. Motivated by a human's ability to adapt, we explore how task plans from a knowledge graph, known as the Functional Object- Oriented Network (FOON), can be generated for novel problems requiring concepts that are not readily available to the robot in its knowledge base. Knowledge from 140 cooking recipes are structured in a FOON knowledge graph, which is used for acquiring task plan sequences known as task trees. Task trees can be modified to replicate recipes in a FOON knowledge graph format, which can be useful for enriching FOON with new recipes containing unknown object and state combinations, by relying upon semantic similarity. We demonstrate the power of task tree generation to create task trees with never-before-seen ingredient and state combinations as seen in recipes from the Recipe1M+ dataset, with which we evaluate the quality of the trees based on how accurately they depict newly added ingredients. Our experimental results show that our system is able to provide task sequences with 76% correctness.
This paper proposes 12 multi-object grasps (MOGs) types from a human and robot grasping data set. The grasp types are then analyzed and organized into a MOG taxonomy. This paper first presents three MOG data collection setups: a human finger tracking setup for multi-object grasping demonstrations, a real system with Barretthand, UR5e arm, and a MOG algorithm, a simulation system with the same settings as the real system. Then the paper describes a novel stochastic grasping routine designed based on a biased random walk to explore the robotic hand's configuration space for feasible MOGs. Based on observations in both the human demonstrations and robotic MOG solutions, this paper proposes 12 MOG types in two groups: shape-based types and function-based types. The new MOG types are compared using six characteristics and then compiled into a taxonomy. This paper then introduces the observed MOG type combinations and shows examples of 16 different combinations.
Deep neural networks (DNNs) often rely on massive labelled data for training, which is inaccessible in many applications. Data augmentation (DA) tackles data scarcity by creating new labelled data from available ones. Different DA methods have different mechanisms and therefore using their generated labelled data for DNN training may help improving DNN's generalisation to different degrees. Combining multiple DA methods, namely multi-DA, for DNN training, provides a way to boost generalisation. Among existing multi-DA based DNN training methods, those relying on knowledge distillation (KD) have received great attention. They leverage knowledge transfer to utilise the labelled data sets created by multiple DA methods instead of directly combining them for training DNNs. However, existing KD-based methods can only utilise certain types of DA methods, incapable of utilising the advantages of arbitrary DA methods. We propose a general multi-DA based DNN training framework capable to use arbitrary DA methods. To train a DNN, our framework replicates a certain portion in the latter part of the DNN into multiple copies, leading to multiple DNNs with shared blocks in their former parts and independent blocks in their latter parts. Each of these DNNs is associated with a unique DA and a newly devised loss that allows comprehensively learning from the data generated by all DA methods and the outputs from all DNNs in an online and adaptive way. The overall loss, i.e., the sum of each DNN's loss, is used for training the DNN. Eventually, one of the DNNs with the best validation performance is chosen for inference. We implement the proposed framework by using three distinct DA methods and apply it for training representative DNNs. Experiments on the popular benchmarks of image classification demonstrate the superiority of our method to several existing single-DA and multi-DA based training methods.
The ever-growing model size and scale of compute have attracted increasing interests in training deep learning models over multiple nodes. However, when it comes to training on cloud clusters, especially across remote clusters, huge challenges are faced. In this work, we introduce a general framework, Nebula-I, for collaboratively training deep learning models over remote heterogeneous clusters, the connections between which are low-bandwidth wide area networks (WANs). We took natural language processing (NLP) as an example to show how Nebula-I works in different training phases that include: a) pre-training a multilingual language model using two remote clusters; and b) fine-tuning a machine translation model using knowledge distilled from pre-trained models, which run through the most popular paradigm of recent deep learning. To balance the accuracy and communication efficiency, in Nebula-I, parameter-efficient training strategies, hybrid parallel computing methods and adaptive communication acceleration techniques are jointly applied. Meanwhile, security strategies are employed to guarantee the safety, reliability and privacy in intra-cluster computation and inter-cluster communication. Nebula-I is implemented with the PaddlePaddle deep learning framework, which can support collaborative training over heterogeneous hardware, e.g. GPU and NPU. Experiments demonstrate that the proposed framework could substantially maximize the training efficiency while preserving satisfactory NLP performance. By using Nebula-I, users can run large-scale training tasks over cloud clusters with minimum developments, and the utility of existed large pre-trained models could be further promoted. We also introduced new state-of-the-art results on cross-lingual natural language inference tasks, which are generated based upon a novel learning framework and Nebula-I.
Relational graph neural networks have garnered particular attention to encode graph context in knowledge graphs (KGs). Although they achieved competitive performance on small KGs, how to efficiently and effectively utilize graph context for large KGs remains an open problem. To this end, we propose the Relation-based Embedding Propagation (REP) method. It is a post-processing technique to adapt pre-trained KG embeddings with graph context. As relations in KGs are directional, we model the incoming head context and the outgoing tail context separately. Accordingly, we design relational context functions with no external parameters. Besides, we use averaging to aggregate context information, making REP more computation-efficient. We theoretically prove that such designs can avoid information distortion during propagation. Extensive experiments also demonstrate that REP has significant scalability while improving or maintaining prediction quality. Notably, it averagely brings about 10% relative improvement to triplet-based embedding methods on OGBL-WikiKG2 and takes 5%-83% time to achieve comparable results as the state-of-the-art GC-OTE.
Given an image with multiple people, our goal is to directly regress the pose and shape of all the people as well as their relative depth. Inferring the depth of a person in an image, however, is fundamentally ambiguous without knowing their height. This is particularly problematic when the scene contains people of very different sizes, e.g. from infants to adults. To solve this, we need several things. First, we develop a novel method to infer the poses and depth of multiple people in a single image. While previous work that estimates multiple people does so by reasoning in the image plane, our method, called BEV, adds an additional imaginary Bird's-Eye-View representation to explicitly reason about depth. BEV reasons simultaneously about body centers in the image and in depth and, by combing these, estimates 3D body position. Unlike prior work, BEV is a single-shot method that is end-to-end differentiable. Second, height varies with age, making it impossible to resolve depth without also estimating the age of people in the image. To do so, we exploit a 3D body model space that lets BEV infer shapes from infants to adults. Third, to train BEV, we need a new dataset. Specifically, we create a "Relative Human" (RH) dataset that includes age labels and relative depth relationships between the people in the images. Extensive experiments on RH and AGORA demonstrate the effectiveness of the model and training scheme. BEV outperforms existing methods on depth reasoning, child shape estimation, and robustness to occlusion. The code and dataset are released for research purposes.
Current face detection algorithms are extremely generalized and can obtain decent accuracy when detecting the adult faces. These approaches are insufficient when handling outlier cases, for example when trying to detect the face of a neonate infant whose face composition and expressions are relatively different than that of the adult. It is furthermore difficult when applied to detect faces in a complicated setting such as the Neonate Intensive Care Unit. By training a state-of-the-art face detection model, You-Only-Look-Once, on a proprietary dataset containing labelled neonate faces in a clinical setting, this work achieves near real time neonate face detection. Our preliminary findings show an accuracy of 68.7%, compared to the off the shelf solution which detected neonate faces with an accuracy of 7.37%. Although further experiments are needed to validate our model, our results are promising and prove the feasibility of detecting neonatal faces in challenging real-world settings. The robust and real-time detection of neonatal faces would benefit wide range of automated systems (e.g., pain recognition and surveillance) who currently suffer from the time and effort due to the necessity of manual annotations. To benefit the research community, we make our trained weights publicly available at github(https://github.com/ja05haus/trained_neonate_face).
Sparse Transformer has recently attracted a lot of attention since the ability for reducing the quadratic dependency on the sequence length. We argue that two factors, information bottleneck sensitivity and inconsistency between different attention topologies, could affect the performance of the Sparse Transformer. This paper proposes a well-designed model named ERNIE-Sparse. It consists of two distinctive parts: (i) Hierarchical Sparse Transformer (HST) to sequentially unify local and global information. (ii) Self-Attention Regularization (SAR) method, a novel regularization designed to minimize the distance for transformers with different attention topologies. To evaluate the effectiveness of ERNIE-Sparse, we perform extensive evaluations. Firstly, we perform experiments on a multi-modal long sequence modeling task benchmark, Long Range Arena (LRA). Experimental results demonstrate that ERNIE-Sparse significantly outperforms a variety of strong baseline methods including the dense attention and other efficient sparse attention methods and achieves improvements by 2.77% (57.78% vs. 55.01%). Secondly, to further show the effectiveness of our method, we pretrain ERNIE-Sparse and verified it on 3 text classification and 2 QA downstream tasks, achieve improvements on classification benchmark by 0.83% (92.46% vs. 91.63%), on QA benchmark by 3.24% (74.67% vs. 71.43%). Experimental results continue to demonstrate its superior performance.
This paper considers the problem of temporal video interpolation, where the goal is to synthesize a new video frame given its two neighbors. We propose Cross-Video Neural Representation (CURE) as the first video interpolation method based on neural fields (NF). NF refers to the recent class of methods for the neural representation of complex 3D scenes that has seen widespread success and application across computer vision. CURE represents the video as a continuous function parameterized by a coordinate-based neural network, whose inputs are the spatiotemporal coordinates and outputs are the corresponding RGB values. CURE introduces a new architecture that conditions the neural network on the input frames for imposing space-time consistency in the synthesized video. This not only improves the final interpolation quality, but also enables CURE to learn a prior across multiple videos. Experimental evaluations show that CURE achieves the state-of-the-art performance on video interpolation on several benchmark datasets.
Double-strand DNA breaks (DSBs) are a form of DNA damage that can cause abnormal chromosomal rearrangements. Recent technologies based on high-throughput experiments have obvious high costs and technical challenges.Therefore, we design a graph neural network based method to predict DSBs (GraphDSB), using DNA sequence features and chromosome structure information. In order to improve the expression ability of the model, we introduce Jumping Knowledge architecture and several effective structural encoding methods. The contribution of structural information to the prediction of DSBs is verified by the experiments on datasets from normal human epidermal keratinocytes (NHEK) and chronic myeloid leukemia cell line (K562), and the ablation studies further demonstrate the effectiveness of the designed components in the proposed GraphDSB framework. Finally, we use GNNExplainer to analyze the contribution of node features and topology to DSBs prediction, and proved the high contribution of 5-mer DNA sequence features and two chromatin interaction modes.
Conventional methods for the image-text generation tasks mainly tackle the naturally bidirectional generation tasks separately, focusing on designing task-specific frameworks to improve the quality and fidelity of the generated samples. Recently, Vision-Language Pre-training models have greatly improved the performance of the image-to-text generation tasks, but large-scale pre-training models for text-to-image synthesis task are still under-developed. In this paper, we propose ERNIE-ViLG, a unified generative pre-training framework for bidirectional image-text generation with transformer model. Based on the image quantization models, we formulate both image generation and text generation as autoregressive generative tasks conditioned on the text/image input. The bidirectional image-text generative modeling eases the semantic alignments across vision and language. For the text-to-image generation process, we further propose an end-to-end training method to jointly learn the visual sequence generator and the image reconstructor. To explore the landscape of large-scale pre-training for bidirectional text-image generation, we train a 10-billion parameter ERNIE-ViLG model on a large-scale dataset of 145 million (Chinese) image-text pairs which achieves state-of-the-art performance for both text-to-image and image-to-text tasks, obtaining an FID of 7.9 on MS-COCO for text-to-image synthesis and best results on COCO-CN and AIC-ICC for image captioning.
Pre-trained language models have achieved state-of-the-art results in various Natural Language Processing (NLP) tasks. GPT-3 has shown that scaling up pre-trained language models can further exploit their enormous potential. A unified framework named ERNIE 3.0 was recently proposed for pre-training large-scale knowledge enhanced models and trained a model with 10 billion parameters. ERNIE 3.0 outperformed the state-of-the-art models on various NLP tasks. In order to explore the performance of scaling up ERNIE 3.0, we train a hundred-billion-parameter model called ERNIE 3.0 Titan with up to 260 billion parameters on the PaddlePaddle platform. Furthermore, we design a self-supervised adversarial loss and a controllable language modeling loss to make ERNIE 3.0 Titan generate credible and controllable texts. To reduce the computation overhead and carbon emission, we propose an online distillation framework for ERNIE 3.0 Titan, where the teacher model will teach students and train itself simultaneously. ERNIE 3.0 Titan is the largest Chinese dense pre-trained model so far. Empirical results show that the ERNIE 3.0 Titan outperforms the state-of-the-art models on 68 NLP datasets.
Transferring multiple objects between bins is a common task for many applications. In robotics, a standard approach is to pick up one object and transfer it at a time. However, grasping and picking up multiple objects and transferring them together at once is more efficient. This paper presents a set of novel strategies for efficiently grasping multiple objects in a bin to transfer them to another. The strategies enable a robotic hand to identify an optimal ready hand configuration (pre-grasp) and calculate a flexion synergy based on the desired quantity of objects to be grasped. This paper also presents an approach that uses the Markov decision process (MDP) to model the pick-transfer routines when the required quantity is larger than the capability of a single grasp. Using the MDP model, the proposed approach can generate an optimal pick-transfer routine that minimizes the number of transfers, representing efficiency. The proposed approach has been evaluated in both a simulation environment and on a real robotic system. The results show the approach reduces the number of transfers by 59% and the number of lifts by 58% compared to an optimal single object pick-transfer solution.
Calorie and nutrition research has attained increased interest in recent years. But, due to the complexity of the problem, literature in this area focuses on a limited subset of ingredients or dish types and simple convolutional neural networks or traditional machine learning. Simultaneously, estimation of ingredient portions can help improve calorie estimation and meal re-production from a given image. In this paper, given a single cooking image, a pipeline for calorie estimation and meal re-production for different servings of the meal is proposed. The pipeline contains two stages. In the first stage, a set of ingredients associated with the meal in the given image are predicted. In the second stage, given image features and ingredients, portions of the ingredients and finally the total meal calorie are simultaneously estimated using a deep transformer-based model. Portion estimation introduced in the model helps improve calorie estimation and is also beneficial for meal re-production in different serving sizes. To demonstrate the benefits of the pipeline, the model can be used for meal kits generation. To evaluate the pipeline, the large scale dataset Recipe1M is used. Prior to experiments, the Recipe1M dataset is parsed and explicitly annotated with portions of ingredients. Experiments show that using ingredients and their portions significantly improves calorie estimation. Also, a visual interface is created in which a user can interact with the pipeline to reach accurate calorie estimations and generate a meal kit for cooking purposes.
Deep neural networks for medical image reconstruction are traditionally trained using high-quality ground-truth images as training targets. Recent work on Noise2Noise (N2N) has shown the potential of using multiple noisy measurements of the same object as an alternative to having a ground-truth. However, existing N2N-based methods are not suitable for learning from the measurements of an object undergoing nonrigid deformation. This paper addresses this issue by proposing the deformation-compensated learning (DeCoLearn) method for training deep reconstruction networks by compensating for object deformations. A key component of DeCoLearn is a deep registration module, which is jointly trained with the deep reconstruction network without any ground-truth supervision. We validate DeCoLearn on both simulated and experimentally collected magnetic resonance imaging (MRI) data and show that it significantly improves imaging quality.
This paper discusses recent research progress in robotic grasping and manipulation in the light of the latest Robotic Grasping and Manipulation Competitions (RGMCs). We first provide an overview of past benchmarks and competitions related to the robotics manipulation field. Then, we discuss the methodology behind designing the manipulation tasks in RGMCs. We provide a detailed analysis of key challenges for each task and identify the most difficult aspects based on the competing teams' performance in recent years. We believe that such an analysis is insightful to determine the future research directions for the robotic manipulation domain.