Meta-Bayesian optimisation (meta-BO) aims to improve the sample efficiency of Bayesian optimisation by leveraging data from related tasks. While previous methods successfully meta-learn either a surrogate model or an acquisition function independently, joint training of both components remains an open challenge. This paper proposes the first end-to-end differentiable meta-BO framework that generalises neural processes to learn acquisition functions via transformer architectures. We enable this end-to-end framework with reinforcement learning (RL) to tackle the lack of labelled acquisition data. Early on, we notice that training transformer-based neural processes from scratch with RL is challenging due to insufficient supervision, especially when rewards are sparse. We formalise this claim with a combinatorial analysis showing that the widely used notion of regret as a reward signal exhibits a logarithmic sparsity pattern in trajectory lengths. To tackle this problem, we augment the RL objective with an auxiliary task that guides part of the architecture to learn a valid probabilistic model as an inductive bias. We demonstrate that our method achieves state-of-the-art regret results against various baselines in experiments on standard hyperparameter optimisation tasks and also outperforms others in the real-world problems of mixed-integer programming tuning, antibody design, and logic synthesis for electronic design automation.

Neural Processes (NPs) have gained attention in meta-learning for their ability to quantify uncertainty, together with their rapid prediction and adaptability. However, traditional NPs are prone to underfitting. Transformer Neural Processes (TNPs) significantly outperform existing NPs, yet their applicability in real-world scenarios is hindered by their quadratic computational complexity relative to both context and target data points. To address this, pseudo-token-based TNPs (PT-TNPs) have emerged as a novel NPs subset that condense context data into latent vectors or pseudo-tokens, reducing computational demands. We introduce the Induced Set Attentive Neural Processes (ISANPs), employing Induced Set Attention and an innovative query phase to improve querying efficiency. Our evaluations show that ISANPs perform competitively with TNPs and often surpass state-of-the-art models in 1D regression, image completion, contextual bandits, and Bayesian optimization. Crucially, ISANPs offer a tunable balance between performance and computational complexity, which scale well to larger datasets where TNPs face limitations.

Stochastic processes model various natural phenomena from disease transmission to stock prices, but simulating and quantifying their uncertainty can be computationally challenging. For example, modeling a Gaussian Process with standard statistical methods incurs an $\mathcal{O}(n^3)$ penalty, and even using state-of-the-art Neural Processes (NPs) incurs an $\mathcal{O}(n^2)$ penalty due to the attention mechanism. We introduce the Transformer Neural Process - Kernel Regression (TNP-KR), a new architecture that incorporates a novel transformer block we call a Kernel Regression Block (KRBlock), which reduces the computational complexity of attention in transformer-based Neural Processes (TNPs) from $\mathcal{O}((n_C+n_T)^2)$ to $O(n_C^2+n_Cn_T)$ by eliminating masked computations, where $n_C$ is the number of context, and $n_T$ is the number of test points, respectively, and a fast attention variant that further reduces all attention calculations to $\mathcal{O}(n_C)$ in space and time complexity. In benchmarks spanning such tasks as meta-regression, Bayesian optimization, and image completion, we demonstrate that the full variant matches the performance of state-of-the-art methods while training faster and scaling two orders of magnitude higher in number of test points, and the fast variant nearly matches that performance while scaling to millions of both test and context points on consumer hardware.

Meta-Bayesian optimisation (meta-BO) aims to improve the sample efficiency of Bayesian optimisation by leveraging data from related tasks. While previous methods successfully meta-learn either a surrogate model or an acquisition function independently, joint training of both components remains an open challenge. This paper proposes the first end-to-end differentiable meta-BO framework that generalises neural processes to learn acquisition functions via transformer architectures. We enable this end-to-end framework with reinforcement learning (RL) to tackle the lack of labelled acquisition data. Early on, we notice that training transformer-based neural processes from scratch with RL is challenging due to insufficient supervision, especially when rewards are sparse.

Neural Processes (NPs) are a popular class of approaches for meta-learning. Similar to Gaussian Processes (GPs), NPs define distributions over functions and can estimate uncertainty in their predictions. However, unlike GPs, NPs and their variants suffer from underfitting and often have intractable likelihoods, which limit their applications in sequential decision making. We propose Transformer Neural Processes (TNPs), a new member of the NP family that casts uncertainty-aware meta learning as a sequence modeling problem. We learn TNPs via an autoregressive likelihood-based objective and instantiate it with a novel transformer-based architecture. The model architecture respects the inductive biases inherent to the problem structure, such as invariance to the observed data points and equivariance to the unobserved points. We further investigate knobs within the TNP framework that tradeoff expressivity of the decoding distribution with extra computation. Empirically, we show that TNPs achieve state-of-the-art performance on various benchmark problems, outperforming all previous NP variants on meta regression, image completion, contextual multi-armed bandits, and Bayesian optimization.