In constrained Markov decision processes (CMDPs) with adversarial rewards and constraints, a well-known impossibility result prevents any algorithm from attaining both sublinear regret and sublinear constraint violation, when competing against a best-in-hindsight policy that satisfies constraints on average. In this paper, we show that this negative result can be eased in CMDPs with non-stationary rewards and constraints, by providing algorithms whose performances smoothly degrade as non-stationarity increases. Specifically, we propose algorithms attaining $\tilde{\mathcal{O}} (\sqrt{T} + C)$ regret and positive constraint violation under bandit feedback, where $C$ is a corruption value measuring the environment non-stationarity. This can be $\Theta(T)$ in the worst case, coherently with the impossibility result for adversarial CMDPs. First, we design an algorithm with the desired guarantees when $C$ is known. Then, in the case $C$ is unknown, we show how to obtain the same results by embedding such an algorithm in a general meta-procedure. This is of independent interest, as it can be applied to any non-stationary constrained online learning setting.

In imperfect information games (e.g. Bridge, Skat, Poker), one of the fundamental considerations is to infer the missing information while at the same time avoiding the disclosure of private information. Disregarding the issue of protecting private information can lead to a highly exploitable performance. Yet, excessive attention to it leads to hesitations that are no longer consistent with our private information. In our work, we show that to improve performance, one must choose whether to use a player's private information. We extend our work by proposing a new belief distribution depending on the amount of private and public information desired. We empirically demonstrate an increase in performance and, with the aim of further improving performance, the new distribution should be used according to the position in the game. Our experiments have been done on multiple benchmarks and in multiple determinization-based algorithms (PIMC and IS-MCTS).

Autonomous navigation in environments is a critical capability for modern mobile robots, and has been extensively studied over several decades. Classical approaches to navigation rely on constructing detailed maps of the environment and accurate localization of the robot within the map [1, 2, 3]. However, with increasing demand for deploying robots in novel uncontrolled environments such as households, last-mile delivery, etc., constructing accurate and fine-grained maps frequently is often impractical. Robots must now be able to navigate without maps, which means efficient navigation policies require accurate semantic understanding of the scene, efficient exploration and episodic memory, and long-horizon planning with limited knowledge of the environment. Advances in scene understanding and a have led to semantic navigation tasks such as image-goal navigation [4, 5], object-goal navigation [6, 7], etc. receiving significant focus in recent years. In this work, we consider the specific task of image-goal navigation where robot's navigation objective is specified by an RGB image. We motivate the task with the following scenario: Consider a mobile household robot equipped with an onboard camera tasked with picking up a novel unseen object (say, a new shirt). Since the robot has no prior knowledge about the novel object, it would need other semantic information to understand the object - an image of the object would serve this purpose effectively.

Predicting simple function classes has been widely used as a testbed for developing theory and understanding of the trained Transformer's in-context learning (ICL) ability. In this paper, we revisit the training of Transformers on linear regression tasks, and different from all the existing literature, we consider a bi-objective prediction task of predicting both the conditional expectation $\mathbb{E}[Y|X]$ and the conditional variance Var$(Y|X)$. This additional uncertainty quantification objective provides a handle to (i) better design out-of-distribution experiments to distinguish ICL from in-weight learning (IWL) and (ii) make a better separation between the algorithms with and without using the prior information of the training distribution. Theoretically, we show that the trained Transformer reaches near Bayes-optimum, suggesting the usage of the information of the training distribution. Our method can be extended to other cases. Specifically, with the Transformer's context window $S$, we prove a generalization bound of $\tilde{\mathcal{O}}(\sqrt{\min\{S, T\}/(n T)})$ on $n$ tasks with sequences of length $T$, providing sharper analysis compared to previous results of $\tilde{\mathcal{O}}(\sqrt{1/n})$. Empirically, we illustrate that while the trained Transformer behaves as the Bayes-optimal solution as a natural consequence of supervised training in distribution, it does not necessarily perform a Bayesian inference when facing task shifts, in contrast to the \textit{equivalence} between these two proposed in many existing literature. We also demonstrate the trained Transformer's ICL ability over covariates shift and prompt-length shift and interpret them as a generalization over a meta distribution.

We resolve the open problem of designing a computationally efficient algorithm for infinite-horizon average-reward linear Markov Decision Processes (MDPs) with $\widetilde{O}(\sqrt{T})$ regret. Previous approaches with $\widetilde{O}(\sqrt{T})$ regret either suffer from computational inefficiency or require strong assumptions on dynamics, such as ergodicity. In this paper, we approximate the average-reward setting by the discounted setting and show that running an optimistic value iteration-based algorithm for learning the discounted setting achieves $\widetilde{O}(\sqrt{T})$ regret when the discounting factor $\gamma$ is tuned appropriately. The challenge in the approximation approach is to get a regret bound with a sharp dependency on the effective horizon $1 / (1 - \gamma)$. We use a computationally efficient clipping operator that constrains the span of the optimistic state value function estimate to achieve a sharp regret bound in terms of the effective horizon, which leads to $\widetilde{O}(\sqrt{T})$ regret.

Continuous normalizing flows (CNFs) learn the probability path between a reference and a target density by modeling the vector field generating said path using neural networks. Recently, Lipman et al. (2022) introduced a simple and inexpensive method for training CNFs in generative modeling, termed flow matching (FM). In this paper, we re-purpose this method for probabilistic inference by incorporating Markovian sampling methods in evaluating the FM objective and using the learned probability path to improve Monte Carlo sampling. We propose a sequential method, which uses samples from a Markov chain to fix the probability path defining the FM objective. We augment this scheme with an adaptive tempering mechanism that allows the discovery of multiple modes in the target. Under mild assumptions, we establish convergence to a local optimum of the FM objective, discuss improvements in the convergence rate, and illustrate our methods on synthetic and real-world examples.

We generalize active learning to address real-world settings with concrete prediction targets where sampling is restricted to an accessible region of the domain, while prediction targets may lie outside this region. We analyze a family of decision rules that sample adaptively to minimize uncertainty about prediction targets. We are the first to show, under general regularity assumptions, that such decision rules converge uniformly to the smallest possible uncertainty obtainable from the accessible data. We demonstrate their strong sample efficiency in two key applications: Active few-shot fine-tuning of large neural networks and safe Bayesian optimization, where they improve significantly upon the state-of-the-art.

In this work, an advanced deep reinforcement learning architecture is used to train neural network agents playing atari games. Given only the raw game pixels, action space, and reward information, the system can train agents to play any Atari game. At first, this system uses advanced techniques like deep Q-networks and dueling Q-networks to train efficient agents, the same techniques used by DeepMind to train agents that beat human players in Atari games. As an extension, plastic neural networks are used as agents, and their feasibility is analyzed in this scenario. The plasticity implementation was based on backpropagation and the Hebbian update rule. Plastic neural networks have excellent features like lifelong learning after the initial training, which makes them highly suitable in adaptive learning environments. As a new analysis of plasticity in this context, this work might provide valuable insights and direction for future works. Einforcement learning is a computational technique where an agent learns by directly interacting with its environment without having a complete model of the environment [1]. Reinforcement learning is a very good example of adaptive systems where an agent learns to make decisions and take actions in an environment in order to maximize some reward, which acts as feedback from the environment to the agent. Well-crafted reinforcement learning agents with optimized training loops are known to learn complex tasks, such as playing computer games. In previous work, a CNN-based agent was trained using discounted policy gradients, where all the rewards in an episode were fed to the agent as training data after discounting by a factor [2]. Although this approach served as a good starting point, it is not suitable for learning to control complex environments, such as Atari games. A better implementation is possible using the Q-Learning algorithm, which is based on the Bellman equation [3]. The Bellman equation is based on the Markov decision process [4] and states that the optimal value of a state is equal to the immediate reward plus the discounted expected optimal value of the next state under the optimal policy. While the Bellman equation requires all the reward values and transition probabilities to be known in advance, the Q-Learning algorithm uses Q-Values, which are initialized as random values and optimized gradually.

Enrollment projection is a critical aspect of university management, guiding decisions related to resource allocation and revenue forecasting. However, despite its importance, there remains a lack of transparency regarding the methodologies utilized by many institutions. This paper presents an innovative approach to enrollment projection using Markov Chain modeling, drawing upon a case study conducted at Eastern Michigan University (EMU). Markov Chain modeling emerges as a promising approach for enrollment projection, offering precise predictions based on historical trends. This paper outlines the implementation of Enhanced Markov Chain modeling at EMU, detailing the methodology used to compute transition probabilities and evaluate model performance. Despite challenges posed by external uncertainties such as the COVID-19 pandemic, Markov Chain modeling has demonstrated impressive accuracy, with an average difference of less than 1 percent between predicted and actual enrollments. The paper concludes with a discussion of future directions and opportunities for collaboration among institutions.

Chain-of-Thought (CoT) reasoning could in principle enable a deeper understanding of a language model's (LM) internal reasoning. However, prior work suggests that LMs can answer questions similarly despite changes in their CoT, suggesting that those models are not truly using the CoT. We propose an reinforcement learning technique to produce CoTs that are sufficient alone for predicting future text, independent of other context. This methodology ensures that if the LM can predict future tokens, then it must have used the CoT to understand its context. We formalize the informativeness of a sender to a receiver LM as the degree to which the sender helps the receiver predict their future observations, and we define a "Markovian" LM as one which predicts future text given only a CoT as context. We derive a "Markovian training" procedure by applying our definition of informativeness to a Markovian LM and optimizing via policy gradient and Proximal Policy Optimization (PPO). We demonstrate our training algorithm's effectiveness on fifteen-term arithmetic problems, show the model utilizes the CoT, and externally validate that the generated CoT is meaningful and usable by another model.