A standard assumption in sequential decision making is that we observe everything required to make good decisions. We discuss two specific examples (temporally correlated noise (a) and unobserved contexts (c)) that have stymied the use of IL/RL algorithms (in autonomous helicopters (b) and self-driving (d)). We derive provably correct algorithms for both of these problems that scale to continuous control problems. Reinforcement Learning (RL) and Imitation Learning (IL) methods have achieved impressive results in recent years like beating the world champion at Go or controlling stratospheric balloons. Usually, these results are on problems where we either a) observe the full state or b) are able to faithfully execute our intended actions on the system.

Offline reinforcement-learning (RL) algorithms learn to make decisions using a given, fixed training dataset without online data collection. This problem setting is captivating because it holds the promise of utilizing previously collected datasets without any costly or risky interaction with the environment. However, this promise also bears the drawback of this setting as the restricted dataset induces uncertainty because the agent can encounter unfamiliar sequences of states and actions that the training data did not cover. To mitigate the destructive uncertainty effects, we need to balance the aspiration to take reward-maximizing actions with the incurred risk due to incorrect ones. In financial economics, modern portfolio theory (MPT) is a method that risk-averse investors can use to construct diversified portfolios that maximize their returns without unacceptable levels of risk. We propose integrating MPT into the agent's decision-making process, presenting a new simple-yet-highly-effective risk-aware planning algorithm for offline RL. Our algorithm allows us to systematically account for the \emph{estimated quality} of specific actions and their \emph{estimated risk} due to the uncertainty. We show that our approach can be coupled with the Transformer architecture to yield a state-of-the-art planner, which maximizes the return for offline RL tasks. Moreover, our algorithm reduces the variance of the results significantly compared to conventional Transformer decoding, which results in a much more stable algorithm -- a property that is essential for the offline RL setting, where real-world exploration and failures can be costly or dangerous.

Real-time bidding (RTB) has become a major paradigm of display advertising. Each ad impression generated from a user visit is auctioned in real time, where demand-side platform (DSP) automatically provides bid price usually relying on the ad impression value estimation and the optimal bid price determination. However, the current bid strategy overlooks large randomness of the user behaviors (e.g., click) and the cost uncertainty caused by the auction competition. In this work, we explicitly factor in the uncertainty of estimated ad impression values and model the risk preference of a DSP under a specific state and market environment via a sequential decision process. Specifically, we propose a novel adaptive risk-aware bidding algorithm with budget constraint via reinforcement learning, which is the first to simultaneously consider estimation uncertainty and the dynamic risk tendency of a DSP. We theoretically unveil the intrinsic relation between the uncertainty and the risk tendency based on value at risk (VaR). Consequently, we propose two instantiations to model risk tendency, including an expert knowledge-based formulation embracing three essential properties and an adaptive learning method based on self-supervised reinforcement learning. We conduct extensive experiments on public datasets and show that the proposed framework outperforms state-of-the-art methods in practical settings.

Almost all multi-agent reinforcement learning algorithms without communication follow the principle of centralized training with decentralized execution. During centralized training, agents can be guided by the same signals, such as the global state. During decentralized execution, however, agents lack the shared signal. Inspired by viewpoint invariance and contrastive learning, we propose consensus learning for cooperative multi-agent reinforcement learning in this paper. Although based on local observations, different agents can infer the same consensus in discrete space. During decentralized execution, we feed the inferred consensus as an explicit input to the network of agents, thereby developing their spirit of cooperation. Our proposed method can be extended to various multi-agent reinforcement learning algorithms with small model changes. Moreover, we carry out them on some fully cooperative tasks and get convincing results.

In many risk-aware and multi-objective reinforcement learning settings, the utility of the user is derived from a single execution of a policy. In these settings, making decisions based on the average future returns is not suitable. For example, in a medical setting a patient may only have one opportunity to treat their illness. Making decisions using just the expected future returns -- known in reinforcement learning as the value -- cannot account for the potential range of adverse or positive outcomes a decision may have. Therefore, we should use the distribution over expected future returns differently to represent the critical information that the agent requires at decision time by taking both the future and accrued returns into consideration. In this paper, we propose two novel Monte Carlo tree search algorithms. Firstly, we present a Monte Carlo tree search algorithm that can compute policies for nonlinear utility functions (NLU-MCTS) by optimising the utility of the different possible returns attainable from individual policy executions, resulting in good policies for both risk-aware and multi-objective settings. Secondly, we propose a distributional Monte Carlo tree search algorithm (DMCTS) which extends NLU-MCTS. DMCTS computes an approximate posterior distribution over the utility of the returns, and utilises Thompson sampling during planning to compute policies in risk-aware and multi-objective settings. Both algorithms outperform the state-of-the-art in multi-objective reinforcement learning for the expected utility of the returns.

Development of guidance, navigation and control frameworks/algorithms for swarms attracted significant attention in recent years. That being said, algorithms for planning swarm allocations/trajectories for engaging with enemy swarms is largely an understudied problem. Although small-scale scenarios can be addressed with tools from differential game theory, existing approaches fail to scale for large-scale multi-agent pursuit evasion (PE) scenarios. In this work, we propose a reinforcement learning (RL) based framework to decompose to large-scale swarm engagement problems into a number of independent multi-agent pursuit-evasion games. We simulate a variety of multi-agent PE scenarios, where finite time capture is guaranteed under certain conditions. The calculated PE statistics are provided as a reward signal to the high level allocation layer, which uses an RL algorithm to allocate controlled swarm units to eliminate enemy swarm units with maximum efficiency. We verify our approach in large-scale swarm-to-swarm engagement simulations.

We study the learning dynamics of self-predictive learning for reinforcement learning, a family of algorithms that learn representations by minimizing the prediction error of their own future latent representations. Despite its recent empirical success, such algorithms have an apparent defect: trivial representations (such as constants) minimize the prediction error, yet it is obviously undesirable to converge to such solutions. Our central insight is that careful designs of the optimization dynamics are critical to learning meaningful representations. We identify that a faster paced optimization of the predictor and semi-gradient updates on the representation, are crucial to preventing the representation collapse. Then in an idealized setup, we show self-predictive learning dynamics carries out spectral decomposition on the state transition matrix, effectively capturing information of the transition dynamics. Building on the theoretical insights, we propose bidirectional self-predictive learning, a novel self-predictive algorithm that learns two representations simultaneously. We examine the robustness of our theoretical insights with a number of small-scale experiments and showcase the promise of the novel representation learning algorithm with large-scale experiments.

We developed a simulator to quantify the effect of changes in environmental parameters on plant growth in precision farming. Our approach combines the processing of plant images with deep convolutional neural networks (CNN), growth curve modeling, and machine learning. As a result, our system is able to predict growth rates based on environmental variables, which opens the door for the development of versatile reinforcement learning agents.

While reinforcement learning (RL) has become a more popular approach for robotics, designing sufficiently informative reward functions for complex tasks has proven to be extremely difficult due their inability to capture human intent and policy exploitation. Preference based RL algorithms seek to overcome these challenges by directly learning reward functions from human feedback. Unfortunately, prior work either requires an unreasonable number of queries implausible for any human to answer or overly restricts the class of reward functions to guarantee the elicitation of the most informative queries, resulting in models that are insufficiently expressive for realistic robotics tasks. Contrary to most works that focus on query selection to \emph{minimize} the amount of data required for learning reward functions, we take an opposite approach: \emph{expanding} the pool of available data by viewing human-in-the-loop RL through the more flexible lens of multi-task learning. Motivated by the success of meta-learning, we pre-train preference models on prior task data and quickly adapt them for new tasks using only a handful of queries. Empirically, we reduce the amount of online feedback needed to train manipulation policies in Meta-World by 20$\times$, and demonstrate the effectiveness of our method on a real Franka Panda Robot. Moreover, this reduction in query-complexity allows us to train robot policies from actual human users. Videos of our results and code can be found at https://sites.google.com/view/few-shot-preference-rl/home.

Valuing mineral assets is a challenging task that is highly dependent on the supply (geological) uncertainty surrounding resources and reserves, and the uncertainty of demand (commodity prices). In this work, a graph-based reasoning, modeling and solution approach is proposed to jointly address mineral asset valuation and mine plan scheduling and optimization under supply and demand uncertainty in the "mining complex" framework. Three graph-based solutions are proposed: (i) a neural branching policy that learns a block-sampling ore body representation, (ii) a guiding policy that learns to explore a heuristic selection tree, (iii) a hyper-heuristic that manages the value/supply chain optimization and dynamics modeled as a graph structure. Results on two large-scale industrial mining complexes show a reduction of up to three orders of magnitude in primal suboptimality, execution time, and number of iterations, and an increase of up to 40% in the mineral asset value.