The pursuit of long-term fairness involves the interplay between decision-making and the underlying data generating process. In this paper, through causal modeling with a directed acyclic graph (DAG) on the decision-distribution interplay, we investigate the possibility of achieving long-term fairness from a dynamic perspective. We propose Tier Balancing, a technically more challenging but more natural notion to achieve in the context of long-term, dynamic fairness analysis. Different from previous fairness notions that are defined purely on observed variables, our notion goes one step further, capturing behind-the-scenes situation changes on the unobserved latent causal factors that directly carry out the influence from the current decision to the future data distribution. Under the specified dynamics, we prove that in general one cannot achieve the long-term fairness goal only through one-step interventions. Furthermore, in the effort of approaching long-term fairness, we consider the mission of "getting closer to" the long-term fairness goal and present possibility and impossibility results accordingly.

Exploiting unmanned aerial vehicles (UAVs) to execute tasks is gaining growing popularity recently. To solve the underlying task scheduling problem, the deep reinforcement learning (DRL) based methods demonstrate notable advantage over the conventional heuristics as they rely less on hand-engineered rules. However, their decision space will become prohibitively huge as the problem scales up, thus deteriorating the computation efficiency. To alleviate this issue, we propose a double-level deep reinforcement learning (DL-DRL) approach based on a divide and conquer framework (DCF), where we decompose the task scheduling of multi-UAV into task allocation and route planning. Particularly, we design an encoder-decoder structured policy network in our upper-level DRL model to allocate the tasks to different UAVs, and we exploit another attention based policy network in our lower-level DRL model to construct the route for each UAV, with the objective to maximize the number of executed tasks given the maximum flight distance of the UAV. To effectively train the two models, we design an interactive training strategy (ITS), which includes pre-training, intensive training and alternate training. Experimental results show that our DL-DRL performs favorably against the learning-based and conventional baselines including the OR-Tools, in terms of solution quality and computation efficiency. We also verify the generalization performance of our approach by applying it to larger sizes of up to 1000 tasks. Moreover, we also show via an ablation study that our ITS can help achieve a balance between the performance and training efficiency.

Machine Learning-as-a-Service (MLaaS) has become a widespread paradigm, making even the most complex machine learning models available for clients via e.g. a pay-per-query principle. This allows users to avoid time-consuming processes of data collection, hyperparameter tuning, and model training. However, by giving their customers access to the (predictions of their) models, MLaaS providers endanger their intellectual property, such as sensitive training data, optimised hyperparameters, or learned model parameters. Adversaries can create a copy of the model with (almost) identical behavior using the the prediction labels only. While many variants of this attack have been described, only scattered defence strategies have been proposed, addressing isolated threats. This raises the necessity for a thorough systematisation of the field of model stealing, to arrive at a comprehensive understanding why these attacks are successful, and how they could be holistically defended against. We address this by categorising and comparing model stealing attacks, assessing their performance, and exploring corresponding defence techniques in different settings. We propose a taxonomy for attack and defence approaches, and provide guidelines on how to select the right attack or defence strategy based on the goal and available resources. Finally, we analyse which defences are rendered less effective by current attack strategies.

Unfortunately, the best policy from a set of many policies such estimates are often inaccurate (Fu et al., 2021). As given only logged data, is crucial for applying an alternative, many works have explored using empirical offline RL in real-world settings. One idea that Bellman errors to perform OMS, but have found them to has been extensively explored is to select policies be poor predictors of value model accuracy (Irpan et al., based on the mean squared Bellman error 2019; Paine et al., 2020). This has led to a belief among (MSBE) of the associated Q-functions. However, many researchers that Bellman errors are not useful for previous work has struggled to obtain adequate OMS (Géron, 2019; Fujimoto et al., 2022). OMS performance with Bellman errors, leading many researchers to abandon the idea. To this end, To this end, we propose a new algorithm, Supervised Bellman we elucidate why previous work has seen pessimistic Validation (SBV), that provides a better proxy for the results with Bellman errors and identify true Bellman errors than empirical Bellman errors. SBV conditions under which OMS algorithms based achieves strong performance on diverse tasks ranging from on Bellman errors will perform well. Moreover, healthcare problems (Klasnja et al., 2015) to Atari games we develop a new estimator of the MSBE that is (Bellemare et al., 2013).

Stock trading is one of the popular ways for financial management. However, the market and the environment of economy is unstable and usually not predictable. Furthermore, engaging in stock trading requires time and effort to analyze, create strategies, and make decisions. It would be convenient and effective if an agent could assist or even do the task of analyzing and modeling the past data and then generate a strategy for autonomous trading. Recently, reinforcement learning has been shown to be robust in various tasks that involve achieving a goal with a decision making strategy based on time-series data. In this project, we have developed a pipeline that simulates the stock trading environment and have trained an agent to automate the stock trading process with deep reinforcement learning methods, including deep Q-learning, deep SARSA, and the policy gradient method. We evaluate our platform during relatively good (before 2021) and bad (2021 - 2022) situations. The stocks we've evaluated on including Google, Apple, Tesla, Meta, Microsoft, and IBM. These stocks are among the popular ones, and the changes in trends are representative in terms of having good and bad situations. We showed that before 2021, the three reinforcement methods we have tried always provide promising profit returns with total annual rates around $70\%$ to $90\%$, while maintain a positive profit return after 2021 with total annual rates around 2% to 7%.

The cost efficiency of model inference is critical to real-world machine learning (ML) applications, especially for delay-sensitive tasks and resource-limited devices. A typical dilemma is: in order to provide complex intelligent services (e.g. smart city), we need inference results of multiple ML models, but the cost budget (e.g. GPU memory) is not enough to run all of them. In this work, we study underlying relationships among black-box ML models and propose a novel learning task: model linking, which aims to bridge the knowledge of different black-box models by learning mappings (dubbed model links) between their output spaces. We propose the design of model links which supports linking heterogeneous black-box ML models. Also, in order to address the distribution discrepancy challenge, we present adaptation and aggregation methods of model links. Based on our proposed model links, we developed a scheduling algorithm, named MLink. Through collaborative multi-model inference enabled by model links, MLink can improve the accuracy of obtained inference results under the cost budget. We evaluated MLink on a multi-modal dataset with seven different ML models and two real-world video analytics systems with six ML models and 3,264 hours of video. Experimental results show that our proposed model links can be effectively built among various black-box models. Under the budget of GPU memory, MLink can save 66.7% inference computations while preserving 94% inference accuracy, which outperforms multi-task learning, deep reinforcement learning-based scheduler and frame filtering baselines.

We study emergent communication in a multi-agent reinforcement learning setting, where the agents solve cooperative tasks and have access to a communication channel. The communication channel may consist of either discrete symbols or continuous variables. We introduce an inductive bias to aid with the emergence of good communication protocols for continuous messages, and we look at the effect this type of inductive bias has for continuous and discrete messages in itself or when used in combination with reinforcement learning. We demonstrate that this type of inductive bias has a beneficial effect on the communication protocols learnt in two toy environments, Negotiation and Sequence Guess.

Online tuning of real-world plants is a complex optimisation problem that continues to require manual intervention by experienced human operators. Autonomous tuning is a rapidly expanding field of research, where learning-based methods, such as Reinforcement Learning-trained Optimisation (RLO) and Bayesian optimisation (BO), hold great promise for achieving outstanding plant performance and reducing tuning times. Which algorithm to choose in different scenarios, however, remains an open question. Here we present a comparative study using a routine task in a real particle accelerator as an example, showing that RLO generally outperforms BO, but is not always the best choice. Based on the study's results, we provide a clear set of criteria to guide the choice of algorithm for a given tuning task. These can ease the adoption of learning-based autonomous tuning solutions to the operation of complex real-world plants, ultimately improving the availability and pushing the limits of operability of these facilities, thereby enabling scientific and engineering advancements.

Offline reinforcement learning (RL) methodologies enforce constraints on the policy to adhere closely to the behavior policy, thereby stabilizing value learning and mitigating the selection of out-of-distribution (OOD) actions during test time. Conventional approaches apply identical constraints for both value learning and test time inference. However, our findings indicate that the constraints suitable for value estimation may in fact be excessively restrictive for action selection during test time. To address this issue, we propose a Mildly Constrained Evaluation Policy (MCEP) for test time inference with a more constrained target policy for value estimation. Since the target policy has been adopted in various prior approaches, MCEP can be seamlessly integrated with them as a plug-in. We instantiate MCEP based on TD3-BC [Fujimoto and Gu, 2021] and AWAC [Nair et al., 2020] algorithms. The empirical results on MuJoCo locomotion tasks show that the MCEP significantly outperforms the target policy and achieves competitive results to state-of-the-art offline RL methods. The codes are open-sourced at https://github.com/egg-west/MCEP.git.

Preference-based Reinforcement Learning (PbRL) has demonstrated remarkable efficacy in aligning rewards with human intentions. However, a significant challenge lies in the need of substantial human labels, which is costly and time-consuming. Additionally, the expensive preference data obtained from prior tasks is not typically reusable for subsequent task learning, leading to extensive labeling for each new task. In this paper, we propose a novel zero-shot preference-based RL algorithm that leverages labeled preference data from source tasks to infer labels for target tasks, eliminating the requirement for human queries. Our approach utilizes Gromov-Wasserstein distance to align trajectory distributions between source and target tasks. The solved optimal transport matrix serves as a correspondence between trajectories of two tasks, making it possible to identify corresponding trajectory pairs between tasks and transfer the preference labels. However, learning directly from inferred labels that contains a fraction of noisy labels will result in an inaccurate reward function, subsequently affecting policy performance. To this end, we introduce Robust Preference Transformer, which models the rewards as Gaussian distributions and incorporates reward uncertainty in addition to reward mean. The empirical results on robotic manipulation tasks of Meta-World and Robomimic show that our method has strong capabilities of transferring preferences between tasks and learns reward functions from noisy labels robustly. Furthermore, we reveal that our method attains near-oracle performance with a small proportion of scripted labels.