Weak labeling is a popular weak supervision strategy for Named Entity Recognition (NER) tasks, with the goal of reducing the necessity for hand-crafted annotations. Although there are numerous remarkable annotation tools for NER labeling, the subject of integrating weak labeling sources is still unexplored. We introduce a web-based tool for text annotation called SciAnnotate, which stands for scientific annotation tool. Compared to frequently used text annotation tools, our annotation tool allows for the development of weak labels in addition to providing a manual annotation experience. Our tool provides users with multiple user-friendly interfaces for creating weak labels. SciAnnotate additionally allows users to incorporate their own language models and visualize the output of their model for evaluation. In this study, we take multi-source weak label denoising as an example, we utilized a Bertifying Conditional Hidden Markov Model to denoise the weak label generated by our tool. We also evaluate our annotation tool against the dataset provided by Mysore which contains 230 annotated materials synthesis procedures. The results shows that a 53.7% reduction in annotation time obtained AND a 1.6\% increase in recall using weak label denoising. Online demo is available at https://sciannotate.azurewebsites.net/(demo account can be found in README), but we don't host a model server with it, please check the README in supplementary material for model server usage.

Many advances in cooperative multi-agent reinforcement learning (MARL) are based on two common design principles: value decomposition and parameter sharing. A typical MARL algorithm of this fashion decomposes a centralized Q-function into local Q-networks with parameters shared across agents. Such an algorithmic paradigm enables centralized training and decentralized execution (CTDE) and leads to efficient learning in practice. Despite all the advantages, we revisit these two principles and show that in certain scenarios, e.g., environments with a highly multi-modal reward landscape, value decomposition, and parameter sharing can be problematic and lead to undesired outcomes. In contrast, policy gradient (PG) methods with individual policies provably converge to an optimal solution in these cases, which partially supports some recent empirical observations that PG can be effective in many MARL testbeds. Inspired by our theoretical analysis, we present practical suggestions on implementing multi-agent PG algorithms for either high rewards or diverse emergent behaviors and empirically validate our findings on a variety of domains, ranging from the simplified matrix and grid-world games to complex benchmarks such as StarCraft Multi-Agent Challenge and Google Research Football. We hope our insights could benefit the community towards developing more general and more powerful MARL algorithms. Check our project website at https://sites.google.com/view/revisiting-marl.

Reinforcement learning aims to learn optimal policies from interaction with environments whose dynamics are unknown. Many methods rely on the approximation of a value function to derive near-optimal policies. In partially observable environments, these functions depend on the complete sequence of observations and past actions, called the history. In this work, we show empirically that recurrent neural networks trained to approximate such value functions internally filter the posterior probability distribution of the current state given the history, called the belief. More precisely, we show that, as a recurrent neural network learns the Q-function, its hidden states become more and more correlated with the beliefs of state variables that are relevant to optimal control. This correlation is measured through their mutual information. In addition, we show that the expected return of an agent increases with the ability of its recurrent architecture to reach a high mutual information between its hidden states and the beliefs. Finally, we show that the mutual information between the hidden states and the beliefs of variables that are irrelevant for optimal control decreases through the learning process. In summary, this work shows that in its hidden states, a recurrent neural network approximating the Q-function of a partially observable environment reproduces a sufficient statistic from the history that is correlated to the relevant part of the belief for taking optimal actions.

We study planning problems faced by robots operating in uncertain environments with incomplete knowledge of state, and actions that are noisy and/or imprecise. This paper identifies a new problem sub-class that models settings in which information is revealed only intermittently through some exogenous process that provides state information periodically. Several practical domains fit this model, including the specific scenario that motivates our research: autonomous navigation of a planetary exploration rover augmented by remote imaging. With an eye to efficient specialized solution methods, we examine the structure of instances of this sub-class. They lead to Markov Decision Processes with exponentially large action-spaces but for which, as those actions comprise sequences of more atomic elements, one may establish performance bounds by comparing policies under different information assumptions. This provides a way in which to construct performance bounds systematically. Such bounds are useful because, in conjunction with the insights they confer, they can be employed in bounding-based methods to obtain high-quality solutions efficiently; the empirical results we present demonstrate their effectiveness for the considered problems. The foregoing has also alluded to the distinctive role that time plays for these problems -- more specifically: time until information is revealed -- and we uncover and discuss several interesting subtleties in this regard.

Path planning is a crucial algorithmic approach for designing robot behaviors. Sampling-based approaches, like rapidly exploring random trees (RRTs) or probabilistic roadmaps, are prominent algorithmic solutions for path planning problems. Despite its exponential convergence rate, RRT can only find suboptimal paths. On the other hand, $\textrm{RRT}^*$, a widely-used extension to RRT, guarantees probabilistic completeness for finding optimal paths but suffers in practice from slow convergence in complex environments. Furthermore, real-world robotic environments are often partially observable or with poorly described dynamics, casting the application of $\textrm{RRT}^*$ in complex tasks suboptimal. This paper studies a novel algorithmic formulation of the popular Monte-Carlo tree search (MCTS) algorithm for robot path planning. Notably, we study Monte-Carlo Path Planning (MCPP) by analyzing and proving, on the one part, its exponential convergence rate to the optimal path in fully observable Markov decision processes (MDPs), and on the other part, its probabilistic completeness for finding feasible paths in partially observable MDPs (POMDPs) assuming limited distance observability (proof sketch). Our algorithmic contribution allows us to employ recently proposed variants of MCTS with different exploration strategies for robot path planning. Our experimental evaluations in simulated 2D and 3D environments with a 7 degrees of freedom (DOF) manipulator, as well as in a real-world robot path planning task, demonstrate the superiority of MCPP in POMDP tasks.

Surface grading, the process of leveling an uneven area containing pre-dumped sand piles, is an important task in the construction site pipeline. This labour-intensive process is often carried out by a dozer, a key machinery tool at any construction site. Current attempts to automate surface grading assume perfect localization. However, in real-world scenarios, this assumption fails, as agents are presented with imperfect perception, which leads to degraded performance. In this work, we address the problem of autonomous grading under uncertainties. First, we implement a simulation and a scaled real-world prototype environment to enable rapid policy exploration and evaluation in this setting. Second, we formalize the problem as a partially observable markov decision process and train an agent capable of handling such uncertainties. We show, through rigorous experiments, that an agent trained under perfect localization will suffer degraded performance when presented with localization uncertainties. However, an agent trained using our method will develop a more robust policy for addressing such errors and, consequently, exhibit a better grading performance.

Reinforcement learning (RL) has been applied to attack graphs for penetration testing, however, trained agents do not reflect reality because the attack graphs lack operational nuances typically captured within the intelligence preparation of the battlefield (IPB) that include notions of (cyber) terrain. In particular, current practice constructs attack graphs exclusively using the Common Vulnerability Scoring System (CVSS) and its components. We present methods for constructing attack graphs using notions from IPB on cyber terrain analysis of obstacles, avenues of approach, key terrain, observation and fields of fire, and cover and concealment. We demonstrate our methods on an example where firewalls are treated as obstacles and represented in (1) the reward space and (2) the state dynamics. We show that terrain analysis can be used to bring realism to attack graphs for RL.

With the development of hardware and algorithms, ASR(Automatic Speech Recognition) systems evolve a lot. As The models get simpler, the difficulty of development and deployment become easier, ASR systems are getting closer to our life. On the one hand, we often use APPs or APIs of ASR to generate subtitles and record meetings. On the other hand, smart speaker and self-driving car rely on ASR systems to control AIoT devices. In past few years, there are a lot of works on adversarial examples attacks against ASR systems. By adding a small perturbation to the waveforms, the recognition results make a big difference. In this paper, we describe the development of ASR system, different assumptions of attacks, and how to evaluate these attacks. Next, we introduce the current works on adversarial examples attacks from two attack assumptions: white-box attack and black-box attack. Different from other surveys, we pay more attention to which layer they perturb waveforms in ASR system, the relationship between these attacks, and their implementation methods. We focus on the effect of their works.

Reinforcement Learning (RL) techniques have drawn great attention in many challenging tasks, but their performance deteriorates dramatically when applied to real-world problems. Various methods, such as domain randomization, have been proposed to deal with such situations by training agents under different environmental setups, and therefore they can be generalized to different environments during deployment. However, they usually do not incorporate the underlying environmental factor information that the agents interact with properly and thus can be overly conservative when facing changes in the surroundings. In this paper, we first formalize the task of adapting to changing environmental dynamics in RL as a generalization problem using Contextual Markov Decision Processes (CMDPs). We then propose the Asymmetric Actor-Critic in Contextual RL (AACC) as an end-to-end actor-critic method to deal with such generalization tasks. We demonstrate the essential improvements in the performance of AACC over existing baselines experimentally in a range of simulated environments.

Autonomous navigation in highly populated areas remains a challenging task for robots because of the difficulty in guaranteeing safe interactions with pedestrians in unstructured situations. In this work, we present a crowd navigation control framework that delivers continuous obstacle avoidance and post-contact control evaluated on an autonomous personal mobility vehicle. We propose evaluation metrics for accounting efficiency, controller response and crowd interactions in natural crowds. We report the results of over 110 trials in different crowd types: sparse, flows, and mixed traffic, with low- (< 0.15 ppsm), mid- (< 0.65 ppsm), and high- (< 1 ppsm) pedestrian densities. We present comparative results between two low-level obstacle avoidance methods and a baseline of shared control. Results show a 10% drop in relative time to goal on the highest density tests, and no other efficiency metric decrease. Moreover, autonomous navigation showed to be comparable to shared-control navigation with a lower relative jerk and significantly higher fluency in commands indicating high compatibility with the crowd. We conclude that the reactive controller fulfils a necessary task of fast and continuous adaptation to crowd navigation, and it should be coupled with high-level planners for environmental and situational awareness.