Recently, neural network compression schemes like channel pruning have been widely used to reduce the model size and computational complexity of deep neural network (DNN) for applications in power-constrained scenarios such as embedded systems. Reinforcement learning (RL)-based auto-pruning has been further proposed to automate the DNN pruning process to avoid expensive hand-crafted work. However, the RL-based pruner involves a time-consuming training process and the high expense of each sample further exacerbates this problem. These impediments have greatly restricted the real-world application of RL-based auto-pruning. Thus, in this paper, we propose an efficient auto-pruning framework which solves this problem by taking advantage of the historical data from the previous auto-pruning process. In our framework, we first boost the convergence of the RL-pruner by transfer learning. Then, an augmented transfer learning scheme is proposed to further speed up the training process by improving the transferability. Finally, an assistant learning process is proposed to improve the sample efficiency of the RL agent. The experiments have shown that our framework can accelerate the auto-pruning process by 1.5-2.5 times for ResNet20, and 1.81-2.375 times for other neural networks like ResNet56, ResNet18, and MobileNet v1.
In this work we present a framework that is capable of accurately representing soft robotic actuators in a multiphysics environment in real-time. We propose a constraint-based dynamics model of a 1-dimensional pneumatic soft actuator that accounts for internal pressure forces, as well as the effect of actuator latency and damping under inflation and deflation and demonstrate its accuracy a full soft robotic snake with the composition of multiple 1D actuators. We verify our model's accuracy in static deformation and dynamic locomotion open-loop control experiments. To achieve real-time performance we leverage the parallel computation power of GPUs to allow interactive control and feedback.
As the computational power of toady's devices increases, real-time physically-based rendering becomes possible, and is rapidly gaining attention across a variety of domains. These include gaming, where physically-based rendering enhances immersion and overall entertainment experience, all the way to medicine, where it constitutes a powerful tool for intuitive volumetric data visualization. However, leveraging the obvious benefits of physically-based rendering (also referred to as photo-realistic rendering) remains challenging on embedded devices such as optical see-through head-mounted displays because of their limited computational power, and restricted memory usage and power consumption. We propose methods that aim at overcoming these limitations, fueling the implementation of real-time physically-based rendering on embedded devices. We navigate the compromise between memory requirement, computational power, and image quality to achieve reasonable rendering results by introducing a flexible representation of plenoptic functions and adapting a fast approximation algorithm for image generation from our plenoptic functions. We conclude by discussing potential applications and limitations of the proposed method.
Grasping in dynamic environments presents a unique set of challenges. A stable and reachable grasp can become unreachable and unstable as the target object moves, motion planning needs to be adaptive and in real time, the delay in computation makes prediction necessary. In this paper, we present a dynamic grasping framework that is reachability-aware and motion-aware. Specifically, we model the reachability space of the robot using a signed distance field which enables us to quickly screen unreachable grasps. Also, we train a neural network to predict the grasp quality conditioned on the current motion of the target. Using these as ranking functions, we quickly filter a large grasp database to a few grasps in real time. In addition, we present a seeding approach for arm motion generation that utilizes solution from previous time step. This quickly generates a new arm trajectory that is close to the previous plan and prevents fluctuation. We implement a recurrent neural network (RNN) for modelling and predicting the object motion. Our extensive experiments demonstrate the importance of each of these components and we validate our pipeline on a real robot.
Microaneurysms (MAs) are one of the earliest signs of diabetic retinopathy (DR), a frequent complication of diabetes that can lead to visual impairment and blindness. Adaptive optics scanning laser ophthalmoscopy (AOSLO) provides real-time retinal images with resolution down to 2 $\mu m$ and thus allows detection of the morphologies of individual MAs, a potential marker that might dictate MA pathology and affect the progression of DR. In contrast to the numerous automatic models developed for assessing the number of MAs on fundus photographs, currently there is no high throughput image protocol available for automatic analysis of AOSLO photographs. To address this urgency, we introduce AOSLO-net, a deep neural network framework with customized training policies to automatically segment MAs from AOSLO images. We evaluate the performance of AOSLO-net using 87 DR AOSLO images and our results demonstrate that the proposed model outperforms the state-of-the-art segmentation model both in accuracy and cost and enables correct MA morphological classification.
Face recognition has achieved significant progress in deep-learning era due to the ultra-large-scale and well-labeled datasets. However, training on ultra-large-scale datasets is time-consuming and takes up a lot of hardware resource. Therefore, designing an efficient training approach is crucial and indispensable. The heavy computational and memory costs mainly result from the high dimensionality of the Fully-Connected (FC) layer. Specifically, the dimensionality is determined by the number of face identities, which can be million-level or even more. To this end, we propose a novel training approach for ultra-large-scale face datasets, termed Faster Face Classification (F$^2$C). In F$^2$C, we first define a Gallery Net and a Probe Net that are used to generate identities' centers and extract faces' features for face recognition, respectively. Gallery Net has the same structure as Probe Net and inherits the parameters from Probe Net with a moving average paradigm. After that, to reduce the training time and hardware costs of the FC layer, we propose a Dynamic Class Pool (DCP) that stores the features from Gallery Net and calculates the inner product (logits) with positive samples (whose identities are in the DCP) in each mini-batch. DCP can be regarded as a substitute for the FC layer but it is far smaller, thus greatly reducing the computational and memory costs. For negative samples (whose identities are not in DCP), we minimize the cosine similarities between negative samples and those in DCP. Then, to improve the update efficiency of DCP's parameters, we design a dual data-loader including identity-based and instance-based loaders to generate a certain of identities and samples in mini-batches.
StarCraft II (SC2) is a real-time strategy game, in which players produce and control multiple units to win. Due to its difficulties, such as huge state space, various action space, a long time horizon, and imperfect information, SC2 has been a research highlight in reinforcement learning research. Recently, an SC2 agent called AlphaStar is proposed which shows excellent performance, obtaining a high win-rates of 99.8% against Grandmaster level human players. We implemented a mini-scaled version of it called mini-AlphaStar based on their paper and the pseudocode they provided. The usage and analysis of it are shown in this technical report. The difference between AlphaStar and mini-AlphaStar is that we substituted the hyper-parameters in the former version with much smaller ones for mini-scale training. The codes of mini-AlphaStar are all open-sourced. The objective of mini-AlphaStar is to provide a reproduction of the original AlphaStar and facilitate the future research of RL on large-scale problems.
Reinforcement learning (RL) studies how an agent comes to achieve reward in an environment through interactions over time. Recent advances in machine RL have surpassed human expertise at the world's oldest board games and many classic video games, but they require vast quantities of experience to learn successfully -- none of today's algorithms account for the human ability to learn so many different tasks, so quickly. Here we propose a new approach to this challenge based on a particularly strong form of model-based RL which we call Theory-Based Reinforcement Learning, because it uses human-like intuitive theories -- rich, abstract, causal models of physical objects, intentional agents, and their interactions -- to explore and model an environment, and plan effectively to achieve task goals. We instantiate the approach in a video game playing agent called EMPA (the Exploring, Modeling, and Planning Agent), which performs Bayesian inference to learn probabilistic generative models expressed as programs for a game-engine simulator, and runs internal simulations over these models to support efficient object-based, relational exploration and heuristic planning. EMPA closely matches human learning efficiency on a suite of 90 challenging Atari-style video games, learning new games in just minutes of game play and generalizing robustly to new game situations and new levels. The model also captures fine-grained structure in people's exploration trajectories and learning dynamics. Its design and behavior suggest a way forward for building more general human-like AI systems.
In coal-fired power plants, it is critical to improve the operational efficiency of boilers for sustainability. In this work, we formulate real-time boiler control as an optimization problem that looks for the best distribution of temperature in different zones and oxygen content from the flue to improve the boiler's stability and energy efficiency. We employ an efficient algorithm by integrating appropriate machine learning and optimization techniques. We obtain a large dataset collected from a real boiler for more than two months from our industry partner, and conduct extensive experiments to demonstrate the effectiveness and efficiency of the proposed algorithm.
Time series data account for a major part of data supply available today. Time series mining handles several tasks such as classification, clustering, query-by-content, prediction, and others. Performing data mining tasks on raw time series is inefficient as these data are high-dimensional by nature. Instead, time series are first pre-processed using several techniques before different data mining tasks can be performed on them. In general, there are two main approaches to reduce time series dimensionality, the first is what we call landmark methods. These methods are based on finding characteristic features in the target time series. The second is based on data transformations. These methods transform the time series from the original space into a reduced space, where they can be managed more efficiently. The method we present in this paper applies a third approach, as it projects a time series onto a lower-dimensional space by selecting important points in the time series. The novelty of our method is that these points are not chosen according to a geometric criterion, which is subjective in most cases, but through an optimization process. The other important characteristic of our method is that these important points are selected on a dataset-level and not on a single time series-level. The direct advantage of this strategy is that the distance defined on the low-dimensional space lower bounds the original distance applied to raw data. This enables us to apply the popular GEMINI algorithm. The promising results of our experiments on a wide variety of time series datasets, using different optimizers, and applied to the two major data mining tasks, validate our new method.