arXiv subjects

Jie Fu

Publications and source records attributed to Jie Fu.

269 records · Page 15Linked to original sources

arXiv2018

Adaptive Bidirectional Backpropagation: Towards Biologically Plausible Error Signal Transmission in Neural Networks

The back-propagation (BP) algorithm has been considered the de-facto method for training deep neural networks. It back-propagates errors from the output layer to the hidden layers in an exact manner using the transpose of the feedforward weights. However, it has been argued that this is not biologically plausible because back-propagating error signals with the exact incoming weights are not considered possible in biological neural systems. In this work, we propose a biologically plausible paradigm of neural architecture based on related literature in neuroscience and asymmetric BP-like methods. Specifically, we propose two bidirectional learning algorithms with trainable feedforward and feedback weights. The feedforward weights are used to relay activations from the inputs to target outputs. The feedback weights pass the error signals from the output layer to the hidden layers. Different from other asymmetric BP-like methods, the feedback weights are also plastic in our framework and are trained to approximate the forward activations. Preliminary results show that our models outperform other asymmetric BP-like methods on the MNIST and the CIFAR-10 datasets.

cs.NE↗

arXiv2018

Male pelvic synthetic CT generation from T1-weighted MRI using 2D and 3D convolutional neural networks

To achieve magnetic resonance (MR)-only radiotherapy, a method needs to be employed to estimate a synthetic CT (sCT) for generating electron density maps and patient positioning reference images. We investigated 2D and 3D convolutional neural network (CNN) methods to generate a male pelvic sCT using a T1-weighted MR image. A retrospective study was performed using CTs and T1-weighted MR images of 20 prostate cancer patients. The proposed 2D CNN model, which contained 27 convolutional layers, was modified from the SegNet for better performance. 3D version of the CNN model was also developed. Both CNN models were trained from scratch to map intensities of T1-weighted MR images to CT Hounsfield Unit (HU) values. Each sCT was generated in a five-fold-cross-validation framework and compared with the corresponding CT using voxel-wise mean absolute error (MAE), and dice similarity coefficient (DSC), recall, and precision for bony structures. Wilcoxon signed-rank tests were performed to evaluate the differences between the both models. The MAE averaged across all patients were 40.5 $\pm$ 5.4 HU and 37.6 $\pm$ 5.1 HU for the 2D and 3D CNN models, respectively. The DSC, recall, and precision of the bony structures were 0.81 $\pm$ 0.04, 0.85 $\pm$ 0.04, and 0.77 $\pm$ 0.09 for the 2D CNN model, and 0.82 $\pm$ 0.04, 0.84 $\pm$ 0.04, and 0.80 $\pm$ 0.08 for the 3D CNN model, respectively. P values of the Wilcoxon signed-rank tests were less than 0.05 except for recall, which was 0.6. The 2D and 3D CNN models generated accurate pelvic sCTs for the 20 patients using T1-weighted MR images. The evaluation metrics and statistical tests indicated that the 3D model was able to generate sCTs with better MAE, bone DSC, and bone precision. The accuracy of the dose calculation and patient positioning using generated sCTs will be tested and compared for the two models in the future.

physics.med-ph↗

arXiv2018

Human-in-the-Loop Synthesis for Partially Observable Markov Decision Processes

We study planning problems where autonomous agents operate inside environments that are subject to uncertainties and not fully observable. Partially observable Markov decision processes (POMDPs) are a natural formal model to capture such problems. Because of the potentially huge or even infinite belief space in POMDPs, synthesis with safety guarantees is, in general, computationally intractable. We propose an approach that aims to circumvent this difficulty: in scenarios that can be partially or fully simulated in a virtual environment, we actively integrate a human user to control an agent. While the user repeatedly tries to safely guide the agent in the simulation, we collect data from the human input. Via behavior cloning, we translate the data into a strategy for the POMDP. The strategy resolves all nondeterminism and non-observability of the POMDP, resulting in a discrete-time Markov chain (MC). The efficient verification of this MC gives quantitative insights into the quality of the inferred human strategy by proving or disproving given system specifications. For the case that the quality of the strategy is not sufficient, we propose a refinement method using counterexamples presented to the human. Experiments show that by including humans into the POMDP verification loop we improve the state of the art by orders of magnitude in terms of scalability.

cs.AI↗

arXiv2017

Inverse Reinforce Learning with Nonparametric Behavior Clustering

Inverse Reinforcement Learning (IRL) is the task of learning a single reward function given a Markov Decision Process (MDP) without defining the reward function, and a set of demonstrations generated by humans/experts. However, in practice, it may be unreasonable to assume that human behaviors can be explained by one reward function since they may be inherently inconsistent. Also, demonstrations may be collected from various users and aggregated to infer and predict user's behaviors. In this paper, we introduce the Non-parametric Behavior Clustering IRL algorithm to simultaneously cluster demonstrations and learn multiple reward functions from demonstrations that may be generated from more than one behaviors. Our method is iterative: It alternates between clustering demonstrations into different behavior clusters and inverse learning the reward functions until convergence. It is built upon the Expectation-Maximization formulation and non-parametric clustering in the IRL setting. Further, to improve the computation efficiency, we remove the need of completely solving multiple IRL problems for multiple clusters during the iteration steps and introduce a resampling technique to avoid generating too many unlikely clusters. We demonstrate the convergence and efficiency of the proposed method through learning multiple driver behaviors from demonstrations generated from a grid-world environment and continuous trajectories collected from autonomous robot cars using the Gazebo robot simulator.

cs.AI↗

arXiv2017

Deep Q-Networks for Accelerating the Training of Deep Neural Networks

In this paper, we propose a principled deep reinforcement learning (RL) approach that is able to accelerate the convergence rate of general deep neural networks (DNNs). With our approach, a deep RL agent (synonym for optimizer in this work) is used to automatically learn policies about how to schedule learning rates during the optimization of a DNN. The state features of the agent are learned from the weight statistics of the optimizee during training. The reward function of this agent is designed to learn policies that minimize the optimizee's training time given a certain performance goal. The actions of the agent correspond to changing the learning rate for the optimizee during training. As far as we know, this is the first attempt to use deep RL to learn how to optimize a large-sized DNN. We perform extensive experiments on a standard benchmark dataset and demonstrate the effectiveness of the policies learned by our approach.

cs.LG↗

arXiv2017

Environment-Independent Task Specifications via GLTL

We propose a new task-specification language for Markov decision processes that is designed to be an improvement over reward functions by being environment independent. The language is a variant of Linear Temporal Logic (LTL) that is extended to probabilistic specifications in a way that permits approximations to be learned in finite time. We provide several small environments that demonstrate the advantages of our geometric LTL (GLTL) language and illustrate how it can be used to specify standard reinforcement-learning tasks straightforwardly.

cs.AI↗

arXiv2016

Importance sampling-based approximate optimal planning and control

In this paper, we propose a sampling-based planning and optimal control method of nonlinear systems under non-differentiable constraints. Motivated by developing scalable planning algorithms, we consider the optimal motion plan to be a feedback controller that can be approximated by a weighted sum of given bases. Given this approximate optimal control formulation, our main contribution is to introduce importance sampling, specifically, model-reference adaptive search algorithm, to iteratively compute the optimal weight parameters, i.e., the weights corresponding to the optimal policy function approximation given chosen bases. The key idea is to perform the search by iteratively estimating a parametrized distribution which converges to a Dirac's Delta that infinitely peaks on the global optimal weights. Then, using this direct policy search, we incorporated trajectory-based verification to ensure that, for a class of nonlinear systems, the obtained policy is not only optimal but robust to bounded disturbances. The correctness and efficiency of the methods are demonstrated through numerical experiments including linear systems with a nonlinear cost function and motion planning for a Dubins car.

eess.SY↗

arXiv2016

Dynamics of a prominence-horn structure during its evaporation in the solar corona

The physical connection among and formation mechanisms of various components of the prominence-horn cavity system remain elusive. Here we present observations of such a system, focusing on a section of the prominence that rises and separates gradually from the main body. This forms a configuration sufficiently simple to yield clues to the above issues. It is characterized by embedding horns, oscillations, and a gradual disappearance of the separated material. The prominence-horn structure exhibits a large amplitude longitudinal oscillation with a period of ~150 minutes and an amplitude of ~30 Mm along the trajectory defined by the concave horn structure. The horns also experience a simultaneous transverse oscillation with a much smaller amplitude (~3 Mm) and shorter period (~10-15 minutes), likely representative of a global mode of the large-scale magnetic structure. The gradual disappearance of the structure indicates that the horn, an observational manifestation of the field-aligned transition region separating the cool and dense prominence from the hot and tenuous corona, is formed due to the heating and diluting process of the central prominence mass, while most previous studies suggest that it is the opposite process, i.e., the cooling and condensation of coronal plasmas, to form the horn. This study also demonstrates how the prominence transports magnetic flux to the upper corona, a process essential for the gradual build-up of pre-eruption magnetic energy.

astro-ph.SR↗

arXiv2016

DrMAD: Distilling Reverse-Mode Automatic Differentiation for Optimizing Hyperparameters of Deep Neural Networks

The performance of deep neural networks is well-known to be sensitive to the setting of their hyperparameters. Recent advances in reverse-mode automatic differentiation allow for optimizing hyperparameters with gradients. The standard way of computing these gradients involves a forward and backward pass of computations. However, the backward pass usually needs to consume unaffordable memory to store all the intermediate variables to exactly reverse the forward training procedure. In this work we propose a simple but effective method, DrMAD, to distill the knowledge of the forward pass into a shortcut path, through which we approximately reverse the training trajectory. Experiments on several image benchmark datasets show that DrMAD is at least 45 times faster and consumes 100 times less memory compared to state-of-the-art methods for optimizing hyperparameters with minimal compromise to its effectiveness. To the best of our knowledge, DrMAD is the first research attempt to make it practical to automatically tune thousands of hyperparameters of deep neural networks. The code can be downloaded from https://github.com/bigaidream-projects/drmad

cs.LG↗

arXiv2015

Optimal Temporal Logic Planning in Probabilistic Semantic Maps

This paper considers robot motion planning under temporal logic constraints in probabilistic maps obtained by semantic simultaneous localization and mapping (SLAM). The uncertainty in a map distribution presents a great challenge for obtaining correctness guarantees with respect to the linear temporal logic (LTL) specification. We show that the problem can be formulated as an optimal control problem in which both the semantic map and the logic formula evaluation are stochastic. Our first contribution is to reduce the stochastic control problem for a subclass of LTL to a deterministic shortest path problem by introducing a confidence parameter $delta$. A robot trajectory obtained from the deterministic problem is guaranteed to have minimum cost and to satisfy the logic specification in the true environment with probability $delta$. Our second contribution is to design an admissible heuristic function that guides the planning in the deterministic problem towards satisfying the temporal logic specification. This allows us to obtain an optimal and very efficient solution using the A* algorithm. The performance and correctness of our approach are demonstrated in a simulated semantic environment using a differential-drive robot.

cs.RO↗

arXiv2015

Computational methods for stochastic control with metric interval temporal logic specifications

This paper studies an optimal control problem for continuous-time stochastic systems subject to reachability objectives specified in a subclass of metric interval temporal logic specifications, a temporal logic with real-time constraints. We propose a probabilistic method for synthesizing an optimal control policy that maximizes the probability of satisfying a specification based on a discrete approximation of the underlying stochastic system. First, we show that the original problem can be formulated as a stochastic optimal control problem in a state space augmented with finite memory and states of some clock variables. Second, we present a numerical method for computing an optimal policy with which the given specification is satisfied with the maximal probability in point-based semantics in the discrete approximation of the underlying system. We show that the policy obtained in the discrete approximation converges to the optimal one for satisfying the specification in the continuous or dense-time semantics as the discretization becomes finer in both state and time. Finally, we illustrate our approach with a robotic motion planning example.

eess.SY↗

arXiv2015

Optimal control in Markov decision processes via distributed optimization

Optimal control synthesis in stochastic systems with respect to quantitative temporal logic constraints can be formulated as linear programming problems. However, centralized synthesis algorithms do not scale to many practical systems. To tackle this issue, we propose a decomposition-based distributed synthesis algorithm. By decomposing a large-scale stochastic system modeled as a Markov decision process into a collection of interacting sub-systems, the original control problem is formulated as a linear programming problem with a sparse constraint matrix, which can be solved through distributed optimization methods. Additionally, we propose a decomposition algorithm which automatically exploits, if exists, the modular structure in a given large-scale system. We illustrate the proposed methods through robotic motion planning examples.

eess.SY↗

arXiv2014

Pareto efficiency in synthesizing shared autonomy policies with temporal logic constraints

In systems in which control authority is shared by an autonomous controller and a human operator, it is important to find solutions that achieve a desirable system performance with a reasonable workload for the human operator. We formulate a shared autonomy system capable of capturing the interaction and switching control between an autonomous controller and a human operator, as well as the evolution of the operator's cognitive state during control execution. To trade-off human's effort and the performance level, e.g., measured by the probability of satisfying the underlying temporal logic specification, a two-stage policy synthesis algorithm is proposed for generating Pareto efficient coordination and control policies with respect to user specified weights. We integrate the Tchebychev scalarization method for multi-objective optimization methods to obtain a better coverage of the set of Pareto efficient solutions than linear scalarization methods.

cs.RO↗

arXiv2014

Integrating active sensing into reactive synthesis with temporal logic constraints under partial observations

We introduce the notion of online reactive planning with sensing actions for systems with temporal logic constraints in partially observable and dynamic environments. With incomplete information on the dynamic environment, reactive controller synthesis amounts to solving a two-player game with partial observations, which has impractically computational complexity. To alleviate the high computational burden, online replanning via sensing actions avoids solving the strategy in the reactive system under partial observations. Instead, we only solve for a strategy that ensures a given temporal logic specification can be satisfied had the system have complete observations of its environment. Such a strategy is then transformed into one which makes control decisions based on the observed sequence of states (of the interacting system and its environment). When the system encounters a belief---a set including all possible hypotheses the system has for the current state---for which the observation-based strategy is undefined, a sequence of sensing actions are triggered, chosen by an active sensing strategy, to reduce the uncertainty in the system's belief. We show that by alternating between the observation-based strategy and the active sensing strategy, under a mild technical assumption of the set of sensors in the system, the given temporal logic specification can be satisfied with probability 1.

eess.SY↗

arXiv2014

Abstractions and sensor design in partial-information, reactive controller synthesis

Automated synthesis of reactive control protocols from temporal logic specifications has recently attracted considerable attention in various applications in, for example, robotic motion planning, network management, and hardware design. An implicit and often unrealistic assumption in this past work is the availability of complete and precise sensing information during the execution of the controllers. In this paper, we use an abstraction procedure for systems with partial observation and propose a formalism to investigate effects of limitations in sensing. The abstraction procedure enables the existing synthesis methods with partial observation to be applicable and efficient for systems with infinite (or finite but large number of) states. This formalism enables us to systematically discover sensing modalities necessary in order to render the underlying synthesis problems feasible. We use counterexamples, which witness unrealizability potentially due to the limitations in sensing and the coarseness in the abstract system, and interpolation-based techniques to refine the model and the sensing modalities, i.e., to identify new sensors to be included, in such synthesis problems. We demonstrate the method on examples from robotic motion planning.

eess.SY↗

arXiv2014

Probably Approximately Correct MDP Learning and Control With Temporal Logic Constraints

We consider synthesis of control policies that maximize the probability of satisfying given temporal logic specifications in unknown, stochastic environments. We model the interaction between the system and its environment as a Markov decision process (MDP) with initially unknown transition probabilities. The solution we develop builds on the so-called model-based probably approximately correct Markov decision process (PAC-MDP) methodology. The algorithm attains an $\varepsilon$-approximately optimal policy with probability $1-δ$ using samples (i.e. observations), time and space that grow polynomially with the size of the MDP, the size of the automaton expressing the temporal logic specification, $\frac{1}{\varepsilon}$, $\frac{1}δ$ and a finite time horizon. In this approach, the system maintains a model of the initially unknown MDP, and constructs a product MDP based on its learned model and the specification automaton that expresses the temporal logic constraints. During execution, the policy is iteratively updated using observation of the transitions taken by the system. The iteration terminates in finitely many steps. With high probability, the resulting policy is such that, for any state, the difference between the probability of satisfying the specification under this policy and the optimal one is within a predefined bound.

eess.SY↗

arXiv2012

Symbolic Planning and Control Using Game Theory and Grammatical Inference

This paper presents an approach that brings together game theory with grammatical inference and discrete abstractions in order to synthesize control strategies for hybrid dynamical systems performing tasks in partially unknown but rule-governed adversarial environments. The combined formulation guarantees that a system specification is met if (a) the true model of the environment is in the class of models inferable from a positive presentation, (b) a characteristic sample is observed, and (c) the task specification is satisfiable given the capabilities of the system (agent) and the environment.

cs.RO↗