Enhancing the expressive capacity of deep learning-based time series models with self-supervised pre-training has become ever-increasingly prevalent in time series classification. Even though numerous efforts have been devoted to developing self-supervised models for time series data, we argue that the current methods are not sufficient to learn optimal time series representations due to solely unidirectional encoding over sparse point-wise input units. In this work, we propose TimeMAE, a novel self-supervised paradigm for learning transferrable time series representations based on transformer networks. The distinct characteristics of the TimeMAE lie in processing each time series into a sequence of non-overlapping sub-series via window-slicing partitioning, followed by random masking strategies over the semantic units of localized sub-series. Such a simple yet effective setting can help us achieve the goal of killing three birds with one stone, i.e., (1) learning enriched contextual representations of time series with a bidirectional encoding scheme; (2) increasing the information density of basic semantic units; (3) efficiently encoding representations of time series using transformer networks. Nevertheless, it is a non-trivial to perform reconstructing task over such a novel formulated modeling paradigm. To solve the discrepancy issue incurred by newly injected masked embeddings, we design a decoupled autoencoder architecture, which learns the representations of visible (unmasked) positions and masked ones with two different encoder modules, respectively. Furthermore, we construct two types of informative targets to accomplish the corresponding pretext tasks. One is to create a tokenizer module that assigns a codeword to each masked region, allowing the masked codeword classification (MCC) task to be completed effectively...
Grasping an object when it is in an ungraspable pose is a challenging task, such as books or other large flat objects placed horizontally on a table. Inspired by human manipulation, we address this problem by pushing the object to the edge of the table and then grasping it from the hanging part. In this paper, we develop a model-free Deep Reinforcement Learning framework to synergize pushing and grasping actions. We first pre-train a Variational Autoencoder to extract high-dimensional features of input scenario images. One Proximal Policy Optimization algorithm with the common reward and sharing layers of Actor-Critic is employed to learn both pushing and grasping actions with high data efficiency. Experiments show that our one network policy can converge 2.5 times faster than the policy using two parallel networks. Moreover, the experiments on unseen objects show that our policy can generalize to the challenging case of objects with curved surfaces and off-center irregularly shaped objects. Lastly, our policy can be transferred to a real robot without fine-tuning by using CycleGAN for domain adaption and outperforms the push-to-wall baseline.
This paper explains the generalization power of a deep neural network (DNN) from the perspective of interactive concepts. Many recent studies have quantified a clear emergence of interactive concepts encoded by the DNN, which have been observed on different DNNs during the learning process. Therefore, in this paper, we investigate the generalization power of each interactive concept, and we use the generalization power of different interactive concepts to explain the generalization power of the entire DNN. Specifically, we define the complexity of each interactive concept. We find that simple concepts can be better generalized to testing data than complex concepts. The DNN with strong generalization power usually learns simple concepts more quickly and encodes fewer complex concepts. More crucially, we discover the detouring dynamics of learning complex concepts, which explain both the high learning difficulty and the low generalization power of complex concepts.
In this paper, we present DA-BEV, an implicit depth learning method for Transformer-based camera-only 3D object detection in bird's eye view (BEV). First, a Depth-Aware Spatial Cross-Attention (DA-SCA) module is proposed to take depth into consideration when querying image features to construct BEV features. Then, to make the BEV feature more depth-aware, we introduce an auxiliary learning task, called Depth-wise Contrastive Learning (DCL), by sampling positive and negative BEV features along each ray that connects an object and a camera. DA-SCA and DCL jointly improve the BEV representation and make it more depth-aware. We show that DA-BEV obtains significant improvement (+2.8 NDS) on nuScenes val under the same setting when compared with the baseline method BEVFormer. DA-BEV also achieves strong results of 60.0 NDS and 51.5mAP on nuScenes test with pre-trained VoVNet-99 as backbone. We will release our code.
Model parallelism is conventionally viewed as a method to scale a single large deep learning model beyond the memory limits of a single device. In this paper, we demonstrate that model parallelism can be additionally used for the statistical multiplexing of multiple devices when serving multiple models, even when a single model can fit into a single device. Our work reveals a fundamental trade-off between the overhead introduced by model parallelism and the opportunity to exploit statistical multiplexing to reduce serving latency in the presence of bursty workloads. We explore the new trade-off space and present a novel serving system, AlpaServe, that determines an efficient strategy for placing and parallelizing collections of large deep learning models across a distributed cluster. Evaluation results on production workloads show that AlpaServe can process requests at up to 10x higher rates or 6x more burstiness while staying within latency constraints for more than 99% of requests.
In this paper, we formulate acoustic howling suppression (AHS) as a supervised learning problem and propose a deep learning approach, called Deep AHS, to address it. Deep AHS is trained in a teacher forcing way which converts the recurrent howling suppression process into an instantaneous speech separation process to simplify the problem and accelerate the model training. The proposed method utilizes properly designed features and trains an attention based recurrent neural network (RNN) to extract the target signal from the microphone recording, thus attenuating the playback signal that may lead to howling. Different training strategies are investigated and a streaming inference method implemented in a recurrent mode used to evaluate the performance of the proposed method for real-time howling suppression. Deep AHS avoids howling detection and intrinsically prohibits howling from happening, allowing for more flexibility in the design of audio systems. Experimental results show the effectiveness of the proposed method for howling suppression under different scenarios.
Fundus image captures rear of an eye, and which has been studied for the diseases identification, classification, segmentation, generation, and biological traits association using handcrafted, conventional, and deep learning methods. In biological traits estimation, most of the studies have been carried out for the age prediction and gender classification with convincing results. However, the current study utilizes the cutting-edge deep learning (DL) algorithms to estimate biological traits in terms of age and gender together with associating traits to retinal visuals. For the traits association, our study embeds aging as the label information into the proposed DL model to learn knowledge about the effected regions with aging. Our proposed DL models, named FAG-Net and FGC-Net, correspondingly estimate biological traits (age and gender) and generates fundus images. FAG-Net can generate multiple variants of an input fundus image given a list of ages as conditions. Our study analyzes fundus images and their corresponding association with biological traits, and predicts of possible spreading of ocular disease on fundus images given age as condition to the generative model. Our proposed models outperform the randomly selected state of-the-art DL models.
The Kalman filter is widely used for addressing acoustic echo cancellation (AEC) problems due to their robustness to double-talk and fast convergence. However, the inability to model nonlinearity and the need to tune control parameters cast limitations on such adaptive filtering algorithms. In this paper, we integrate the frequency domain Kalman filter (FDKF) and deep neural networks (DNNs) into a hybrid method, called NeuralKalman, to leverage the advantages of deep learning and adaptive filtering algorithms. Specifically, we employ a DNN to estimate nonlinearly distorted far-end signals, a transition factor, and the nonlinear transition function in the state equation of the FDKF algorithm. Experimental results show that the proposed NeuralKalman improves the performance of FDKF significantly and outperforms strong baseline methods.
The Kalman filter is widely used for addressing acoustic echo cancellation (AEC) problems due to their robustness to double-talk and fast convergence. However, the inability to model nonlinearity and the need to tune control parameters cast limitations on such adaptive filtering algorithms. In this paper, we integrate the frequency domain Kalman filter (FDKF) and deep neural networks (DNNs) into a hybrid method, called KalmanNet, to leverage the advantages of deep learning and adaptive filtering algorithms. Specifically, we employ a DNN to estimate nonlinearly distorted far-end signals, a transition factor, and the nonlinear transition function in the state equation of the FDKF algorithm. Experimental results show that the proposed KalmanNet improves the performance of FDKF significantly and outperforms strong baseline methods.
We present DualCSG, a novel neural network composed of two dual and complementary branches for unsupervised learning of constructive solid geometry (CSG) representations of 3D CAD shapes. Our network is trained to reconstruct a given 3D CAD shape through a compact assembly of quadric surface primitives via fixed-order CSG operations along two branches. The key difference between our method and all previous neural CSG models is that DualCSG has a dedicated branch, the residual branch, to assemble the potentially complex, complement or residual shape that is to be subtracted from an overall cover shape. The cover shape is modeled by the other branch, the cover branch. Both branches construct a union of primitive intersections, where the only difference is that the residual branch also learns primitive inverses while operating in the complement space. With the shape complements, our network is provably general. We demonstrate both quantitatively and qualitatively that our network produces CSG reconstructions with superior quality, more natural trees, and better quality-compactness tradeoff than all existing alternatives, especially over complex and high-genus CAD shapes.