Deep reinforcement learning (RL) provides powerful methods for training optimal sequential decision-making agents. As collecting real-world interactions can entail additional costs and safety risks, the common paradigm of sim2real conducts training in a simulator, followed by real-world deployment. Unfortunately, RL agents easily overfit to the choice of simulated training environments, and worse still, learning ends when the agent masters the specific set of simulated environments. In contrast, the real world is highly open-ended, featuring endlessly evolving environments and challenges, making such RL approaches unsuitable. Simply randomizing over simulated environments is insufficient, as it requires making arbitrary distributional assumptions and can be combinatorially less likely to sample specific environment instances that are useful for learning. An ideal learning process should automatically adapt the training environment to maximize the learning potential of the agent over an open-ended task space that matches or surpasses the complexity of the real world. This thesis develops a class of methods called Unsupervised Environment Design (UED), which aim to produce such open-ended processes. Given an environment design space, UED automatically generates an infinite sequence or curriculum of training environments at the frontier of the learning agent's capabilities. Through extensive empirical studies and theoretical arguments founded on minimax-regret decision theory and game theory, the findings in this thesis show that UED autocurricula can produce RL agents exhibiting significantly improved robustness and generalization to previously unseen environment instances. Such autocurricula are promising paths toward open-ended learning systems that achieve more general intelligence by continually generating and mastering additional challenges of their own design.

Highly effective, task-specific prompts are often heavily engineered by experts to integrate detailed instructions and domain insights based on a deep understanding of both instincts of large language models (LLMs) and the intricacies of the target task. However, automating the generation of such expert-level prompts remains elusive. Existing prompt optimization methods tend to overlook the depth of domain knowledge and struggle to efficiently explore the vast space of expert-level prompts. Addressing this, we present PromptAgent, an optimization method that autonomously crafts prompts equivalent in quality to those handcrafted by experts. At its core, PromptAgent views prompt optimization as a strategic planning problem and employs a principled planning algorithm, rooted in Monte Carlo tree search, to strategically navigate the expert-level prompt space. Inspired by human-like trial-and-error exploration, PromptAgent induces precise expert-level insights and in-depth instructions by reflecting on model errors and generating constructive error feedback. Such a novel framework allows the agent to iteratively examine intermediate prompts (states), refine them based on error feedbacks (actions), simulate future rewards, and search for high-reward paths leading to expert prompts. We apply PromptAgent to 12 tasks spanning three practical domains: BIG-Bench Hard (BBH), as well as domain-specific and general NLP tasks, showing it significantly outperforms strong Chain-of-Thought and recent prompt optimization baselines. Extensive analyses emphasize its capability to craft expert-level, detailed, and domain-insightful prompts with great efficiency and generalizability.

In many text-generation problems, users may prefer not only a single response, but a diverse range of high-quality outputs from which to choose. Quality-diversity (QD) search algorithms aim at such outcomes, by continually improving and diversifying a population of candidates. However, the applicability of QD to qualitative domains, like creative writing, has been limited by the difficulty of algorithmically specifying measures of quality and diversity. Interestingly, recent developments in language models (LMs) have enabled guiding search through AI feedback, wherein LMs are prompted in natural language to evaluate qualitative aspects of text. Leveraging this development, we introduce Quality-Diversity through AI Feedback (QDAIF), wherein an evolutionary algorithm applies LMs to both generate variation and evaluate the quality and diversity of candidate text. When assessed on creative writing domains, QDAIF covers more of a specified search space with high-quality samples than do non-QD controls. Further, human evaluation of QDAIF-generated creative texts validates reasonable agreement between AI and human evaluation. Our results thus highlight the potential of AI feedback to guide open-ended search for creative and original solutions, providing a recipe that seemingly generalizes to many domains and modalities. In this way, QDAIF is a step towards AI systems that can independently search, diversify, evaluate, and improve, which are among the core skills underlying human society's capacity for innovation.

Calibrating agent-based models (ABMs) in economics and finance typically involves a derivative-free search in a very large parameter space. In this work, we benchmark a number of search methods in the calibration of a well-known macroeconomic ABM on real data, and further assess the performance of "mixed strategies" made by combining different methods. We find that methods based on random-forest surrogates are particularly efficient, and that combining search methods generally increases performance since the biases of any single method are mitigated. Moving from these observations, we propose a reinforcement learning (RL) scheme to automatically select and combine search methods on-the-fly during a calibration run. The RL agent keeps exploiting a specific method only as long as this keeps performing well, but explores new strategies when the specific method reaches a performance plateau. The resulting RL search scheme outperforms any other method or method combination tested, and does not rely on any prior information or trial and error procedure.

Self-driving laboratories (SDLs) consist of multiple stations that perform material synthesis and characterisation tasks. To minimize station downtime and maximize experimental throughput, it is practical to run experiments in asynchronous parallel, in which multiple experiments are being performed at once in different stages. Asynchronous parallelization of experiments, however, introduces delayed feedback (i.e. "pending experiments"), which is known to reduce Bayesian optimiser performance. Here, we build a simulator for a multi-stage SDL and compare optimisation strategies for dealing with delayed feedback and asynchronous parallelized operation. Using data from a real SDL, we build a ground truth Bayesian optimisation simulator from 177 previously run experiments for maximizing the conductivity of functional coatings. We then compare search strategies such as expected improvement, noisy expected improvement, 4-mode exploration and random sampling. We evaluate their performance in terms of amount of delay and problem dimensionality. Our simulation results showcase the trade-off between the asynchronous parallel operation and delayed feedback.

We explore the possibility of improving probabilistic models in structured prediction. Specifically, we combine the models with constrained decoding approaches in the context of token classification for information extraction. The decoding methods search for constraint-satisfying label-assignments while maximizing the total probability. To do this, we evaluate several existing approaches, as well as propose a novel decoding method called Lazy-$k$. Our findings demonstrate that constrained decoding approaches can significantly improve the models' performances, especially when using smaller models. The Lazy-$k$ approach allows for more flexibility between decoding time and accuracy. The code for using Lazy-$k$ decoding can be found here: https://github.com/ArthurDevNL/lazyk.

Much of the previous work towards digital agents for graphical user interfaces (GUIs) has relied on text-based representations (derived from HTML or other structured data sources), which are not always readily available. These input representations have been often coupled with custom, task-specific action spaces. This paper focuses on creating agents that interact with the digital world using the same conceptual interface that humans commonly use -- via pixel-based screenshots and a generic action space corresponding to keyboard and mouse actions. Building upon recent progress in pixel-based pretraining, we show, for the first time, that it is possible for such agents to outperform human crowdworkers on the MiniWob++ benchmark of GUI-based instruction following tasks.

Precise hardware performance models play a crucial role in code optimizations. They can assist compilers in making heuristic decisions or aid autotuners in identifying the optimal configuration for a given program. For example, the autotuner for XLA, a machine learning compiler, discovered 10-20% speedup on state-of-the-art models serving substantial production traffic at Google. Although there exist a few datasets for program performance prediction, they target small sub-programs such as basic blocks or kernels. This paper introduces TpuGraphs, a performance prediction dataset on full tensor programs, represented as computational graphs, running on Tensor Processing Units (TPUs). Each graph in the dataset represents the main computation of a machine learning workload, e.g., a training epoch or an inference step. Each data sample contains a computational graph, a compilation configuration, and the execution time of the graph when compiled with the configuration. The graphs in the dataset are collected from open-source machine learning programs, featuring popular model architectures, e.g., ResNet, EfficientNet, Mask R-CNN, and Transformer. TpuGraphs provides 25x more graphs than the largest graph property prediction dataset (with comparable graph sizes), and 770x larger graphs on average compared to existing performance prediction datasets on machine learning programs. This graph-level prediction task on large graphs introduces new challenges in learning, ranging from scalability, training efficiency, to model quality.

While large language models (LLMs) have demonstrated impressive performance on a range of decision-making tasks, they rely on simple acting processes and fall short of broad deployment as autonomous agents. We introduce LATS (Language Agent Tree Search), a general framework that synergizes the capabilities of LLMs in planning, acting, and reasoning. Drawing inspiration from Monte Carlo tree search in model-based reinforcement learning, LATS employs LLMs as agents, value functions, and optimizers, repurposing their latent strengths for enhanced decision-making. What is crucial in this method is the use of an environment for external feedback, which offers a more deliberate and adaptive problem-solving mechanism that moves beyond the limitations of existing techniques. Our experimental evaluation across diverse domains, such as programming, HotPotQA, and WebShop, illustrates the applicability of LATS for both reasoning and acting. In particular, LATS achieves 94.4% for programming on HumanEval with GPT-4 and an average score of 75.9 for web browsing on WebShop with GPT-3.5, demonstrating the effectiveness and generality of our method.

Mobile robots in dynamic environments require fast planning, especially when onboard computational resources are limited. While classic potential field based algorithms may suffice in simple scenarios, in most cases algorithms able to escape local minima are necessary. Configuration-space search algorithms have proven to provide a good trade-off between quality of the solutions and search time. Literature presents a wide variety of approaches that speed up this search by reducing the number of edges that need to be inspected. Much less attention was instead given to reducing the time necessary to evaluate the cost of a single edge. This paper addresses this point by associating edges to motion primitives that prioritize fast collision detection. We show how biarcs can be used as motion primitives that enable fast collision detection, while still providing smooth, tangent continuous paths. The proposed approach does not assume a disc shaped hitbox, making it appealing for all robots with very different width and length or for differential drive robots with active wheels located far from the robot's center.