Recent Semi-Supervised Object Detection (SS-OD) methods are mainly based on self-training, i.e., generating hard pseudo-labels by a teacher model on unlabeled data as supervisory signals. Although they achieved certain success, the limited labeled data in semi-supervised learning scales up the challenges of object detection. We analyze the challenges these methods meet with the empirical experiment results. We find that the massive False Negative samples and inferior localization precision lack consideration. Besides, the large variance of object sizes and class imbalance (i.e., the extreme ratio between background and object) hinder the performance of prior arts. Further, we overcome these challenges by introducing a novel approach, Scale-Equivalent Distillation (SED), which is a simple yet effective end-to-end knowledge distillation framework robust to large object size variance and class imbalance. SED has several appealing benefits compared to the previous works. (1) SED imposes a consistency regularization to handle the large scale variance problem. (2) SED alleviates the noise problem from the False Negative samples and inferior localization precision. (3) A re-weighting strategy can implicitly screen the potential foreground regions of the unlabeled data to reduce the effect of class imbalance. Extensive experiments show that SED consistently outperforms the recent state-of-the-art methods on different datasets with significant margins. For example, it surpasses the supervised counterpart by more than 10 mAP when using 5% and 10% labeled data on MS-COCO.
This letter deals with the problem of clutter edge detection and localization in training data. To this end, the problem is formulated as a binary hypothesis test assuming that the ranks of the clutter covariance matrix are known, and adaptive architectures are designed based on the generalized likelihood ratio test to decide whether the training data within a sliding window contains a homogeneous set or two heterogeneous subsets. In the design stage, we utilize four different covariance matrix structures (i.e., Hermitian, persymmetric, symmetric, and centrosymmetric) to exploit the a priori information. Then, for the case of unknown ranks, the architectures are extended by devising a preliminary estimation stage resorting to the model order selection rules. Numerical examples based on both synthetic and real data highlight that the proposed solutions possess superior detection and localization performance with respect to the competitors that do not use any a priori information.
This paper presents a vision-based modularized drone racing navigation system that uses a customized convolutional neural network (CNN) for the perception module to produce high-level navigation commands and then leverages a state-of-the-art planner and controller to generate low-level control commands, thus exploiting the advantages of both data-based and model-based approaches. Unlike the state-of-the-art method which only takes the current camera image as the CNN input, we further add the latest three drone states as part of the inputs. Our method outperforms the state-of-the-art method in various track layouts and offers two switchable navigation behaviors with a single trained network. The CNN-based perception module is trained to imitate an expert policy that automatically generates ground truth navigation commands based on the pre-computed global trajectories. Owing to the extensive randomization and our modified dataset aggregation (DAgger) policy during data collection, our navigation system, which is purely trained in simulation with synthetic textures, successfully operates in environments with randomly-chosen photorealistic textures without further fine-tuning.
The size of deep neural networks (DNNs) grows rapidly as the complexity of the machine learning algorithm increases. To satisfy the requirement of computation and memory of DNN training, distributed deep learning based on model parallelism has been widely recognized. We propose a new pipeline parallelism training framework, BaPipe, which can automatically explore pipeline parallelism training methods and balanced partition strategies for DNN distributed training. In BaPipe, each accelerator calculates the forward propagation and backward propagation of different parts of networks to implement the intra-batch pipeline parallelism strategy. BaPipe uses a new load balancing automatic exploration strategy that considers the parameters of DNN models and the computation, memory, and communication resources of accelerator clusters. We have trained different DNNs such as VGG-16, ResNet-50, and GNMT on GPU clusters and simulated the performance of different FPGA clusters. Compared with state-of-the-art data parallelism and pipeline parallelism frameworks, BaPipe provides up to 3.2x speedup and 4x memory reduction in various platforms.
The recent development of deep learning has mostly been focusing on Euclidean data, such as images, videos, and audios. However, most real-world information and relationships are often expressed in graphs. Graph convolutional networks (GCNs) appear as a promising approach to efficiently learn from graph data structures, showing advantages in several practical applications such as social network analysis, knowledge discovery, 3D modeling, and motion capturing. However, practical graphs are often extremely large and unbalanced, posting significant performance demand and design challenges on the hardware dedicated to GCN inference. In this paper, we propose an architecture design called Ultra-Workload-Balanced-GCN (UWB-GCN) to accelerate graph convolutional network inference. To tackle the major performance bottleneck of workload imbalance, we propose two techniques: dynamic local sharing and dynamic remote switching, both of which rely on hardware flexibility to achieve performance auto-tuning with negligible area or delay overhead. Specifically, UWB-GCN is able to effectively profile the sparse graph pattern while continuously adjusting the workload distribution among parallel processing elements (PEs). After converging, the ideal configuration is reused for the remaining iterations. To the best of our knowledge, this is the first accelerator design targeted to GCNs and the first work that auto-tunes workload balance in accelerator at runtime through hardware, rather than software, approaches. Our methods can achieve near-ideal workload balance in processing sparse matrices. Experimental results show that UWB-GCN can finish the inference of the Nell graph (66K vertices, 266K edges) in 8.4ms, corresponding to 192x, 289x, and 7.3x respectively, compared to the CPU, GPU, and the baseline GCN design without workload rebalancing.
We present a training pipeline for the autonomous driving task given the current camera image and vehicle speed as the input to produce the throttle, brake, and steering control output. The simulator Airsim's convenient weather and lighting API provides a sufficient diversity during training which can be very helpful to increase the trained policy's robustness. In order to not limit the possible policy's performance, we use a continuous and deterministic control policy setting. We utilize ResNet-34 as our actor and critic networks with some slight changes in the fully connected layers. Considering human's mastery of this task and the high-complexity nature of this task, we first use imitation learning to mimic the given human policy and leverage the trained policy and its weights to the reinforcement learning phase for which we use DDPG. This combination shows a considerable performance boost comparing to both pure imitation learning and pure DDPG for the autonomous driving task.
Deep Neural Networks (DNNs) have revolutionized numerous applications, but the demand for ever more performance remains unabated. Scaling DNN computations to larger clusters is generally done by distributing tasks in batch mode using methods such as distributed synchronous SGD. Among the issues with this approach is that to make the distributed cluster work with high utilization, the workload distributed to each node must be large, which implies nontrivial growth in the SGD mini-batch size. In this paper, we propose a framework called FPDeep, which uses a hybrid of model and layer parallelism to configure distributed reconfigurable clusters to train DNNs. This approach has numerous benefits. First, the design does not suffer from batch size growth. Second, novel workload and weight partitioning leads to balanced loads of both among nodes. And third, the entire system is a fine-grained pipeline. This leads to high parallelism and utilization and also minimizes the time features need to be cached while waiting for back-propagation. As a result, storage demand is reduced to the point where only on-chip memory is used for the convolution layers. We evaluate FPDeep with the Alexnet, VGG-16, and VGG-19 benchmarks. Experimental results show that FPDeep has good scalability to a large number of FPGAs, with the limiting factor being the FPGA-to-FPGA bandwidth. With 6 transceivers per FPGA, FPDeep shows linearity up to 83 FPGAs. Energy efficiency is evaluated with respect to GOPs/J. FPDeep provides, on average, 6.36x higher energy efficiency than comparable GPU servers.
Orthogonal frequency division multiplexing (OFDM) is one of the key technologies that are widely applied in current communication systems. Recently, artificial intelligence (AI)-aided OFDM receivers have been brought to the forefront to break the bottleneck of the traditional OFDM systems. In this paper, we investigate two AI-aided OFDM receivers, data-driven fully connected-deep neural network (FC-DNN) receiver and model-driven ComNet receiver, respectively. We first study their performance under different channel models through simulation and then establish a real-time video transmission system using a 5G rapid prototyping (RaPro) system for over-the-air (OTA) test. To address the performance gap between the simulation and the OTA test caused by the discrepancy between the channel model for offline training and real environments, we develop a novel online training strategy, called SwitchNet receiver. The SwitchNet receiver is with a flexible and extendable architecture and can adapts to real channel by training one parameter online. The OTA test verifies its feasibility and robustness to real environments and indicates its potential for future communications systems. At the end of this paper, we discuss some challenges to inspire future research.