amortised bayesian design

Existing Bayesian experimental design (BED) methods focus primarily on parameter estimation and thus fail to guarantee optimality for downstream decision-making tasks—such as clinical diagnosis or product pricing. This paper proposes a decision-aware amortized BED framework that, for the first time, directly aligns experimental design objectives with decision utility maximization. Our approach features three key contributions: (1) the first end-to-end amortized decision-aware BED model; (2) a novel Transformer-based Neural Decision Process (TNDP) architecture that jointly optimizes experiment selection and decision inference; and (3) an integrated objective combining decision-utility-driven loss with amortized variational inference, enabling end-to-end training of the policy network. Experiments across multiple tasks demonstrate significant improvements in decision accuracy and information efficiency, real-time design speed, and superior performance over conventional BED and state-of-the-art amortized methods.

To address the low sample efficiency and inadequate multimodal coverage in approximate inference for hard-to-sample unnormalized density distributions, this paper reformulates sampling as a sequential decision-making process and introduces a novel adaptive teacher-guided framework: dynamically identifying high-loss regions where the student sampler underperforms and actively constructing a progressive training curriculum. The method integrates reinforcement learning–based normalizing flows, off-policy training, auxiliary behavioral modeling, and amortized inference. Evaluated across synthetic exploration environments, two diffusion-based sampling tasks, and four biochemical discovery benchmarks, it achieves substantial improvements—averaging +37% in sample efficiency and +52% in mode coverage—while notably enhancing discovery of low-probability, high-reward modes.

This work addresses the limited adaptability of Bayesian experimental design under dynamic constraints—such as budget, cost, or physical limitations—by introducing a novel approach that integrates offline amortized inference with online multi-step lookahead planning. The method uniquely combines a pretrained amortized posterior policy with scenario-tree-based online planning to efficiently optimize sequences of experiments while respecting evolving constraints. By jointly leveraging amortized Bayesian inference, scenario tree construction, and constrained optimization, the proposed framework substantially enhances the information gain of selected experiments across diverse constrained tasks, achieving high efficiency and strong adaptability with only modest additional computational overhead.

Bayesian inference often faces a trade-off between computational efficiency and posterior accuracy, especially across multiple datasets. This paper proposes an adaptive hybrid inference workflow that—uniquely—integrates amortized variational inference (AVI) with Markov chain Monte Carlo (MCMC) in a dynamically coordinated manner. Leveraging principled posterior diagnostics, it constructs a Pareto frontier to enable automatic, optimal switching between AVI and MCMC. Computational reuse and scheduling optimization further boost inference throughput. The method unifies generative neural network modeling, MCMC refinement, and verifiable diagnostic mechanisms. Evaluated on tens of thousands of real and synthetic datasets, it achieves a 3.2× average speedup over standalone AVI or MCMC baselines, while preserving posterior fidelity—reducing KL divergence by 47% and increasing effective sample size (ESS) by 2.8×. This work delivers a scalable, efficient, and trustworthy solution for large-scale Bayesian inference.

Repeatedly solving similar optimization problems incurs substantial computational overhead. Method: This paper proposes amortized optimization—a learning-based paradigm that predicts approximate solutions for new problem instances by leveraging structural information from historical problems, thereby accelerating optimization. We establish the first unified theoretical framework for amortized optimization, encompassing variational inference, meta-learning, and optimal transport. Our approach integrates deep neural networks, variational inference, gradient-based meta-learning, and convex optimization modeling to enable end-to-end differentiable optimizer design. Contribution/Results: The framework unifies diverse inference and learning tasks under a coherent theoretical lens; it introduces a general design paradigm for differentiable optimizers; and empirical evaluation demonstrates speedups of several orders of magnitude over conventional optimizers in variational inference, reinforcement learning, and sparse coding—while maintaining strong generalization across tasks.

This work addresses the suboptimal performance of large language models (LLMs) in multi-turn active information-gathering tasks, often due to inefficient questioning strategies. The authors propose ASIG, a novel approach that, for the first time, amortizes Bayesian experimental design (BED) within LLMs by leveraging expected information gain as a reward signal and fine-tuning the model with grouped relative policy optimization. ASIG substantially enhances the model’s active reasoning capabilities: on the 20 Questions task, a 7B-parameter model achieves more than a two-fold increase in success rate while reducing inference cost by over 25×. Furthermore, the method demonstrates strong generalization on the unseen MediQ medical diagnosis benchmark, establishing a highly efficient and transferable framework for multi-turn information acquisition.

Existing adaptive data acquisition methods often suffer from inefficient policy learning due to reliance on biased posterior approximations or inadequate exploitation of model representations. This work proposes POLAR, a novel framework that decouples representation learning from policy learning by leveraging pretrained predictive foundation models to encode belief states. POLAR unifies Bayesian experimental design, Bayesian optimization, and active learning within a single coherent framework. By integrating amortized policy learning with task-specific utility functions, the approach substantially reduces the required number of training samples and consistently outperforms state-of-the-art amortized methods across diverse tasks, significantly enhancing the scalability and efficiency of data acquisition.

This work addresses the challenge of efficiently sampling combinatorial discrete objects from unnormalized posterior distributions, where existing amortized inference methods based on Markov decision processes suffer from state aliasing, leading to impaired signal propagation and limited expressivity. To overcome these limitations, the authors propose a path-dependent amortized sampling framework that introduces a learnable implicit dynamical system, enabling the policy to model the full generation trajectory rather than relying solely on the current state. This approach effectively relaxes the Markov assumption and allows for conditional modeling over entire trajectories. Theoretically, the framework preserves the scalability of existing discrete amortized algorithms under this extended setting. Empirical results demonstrate that the proposed method significantly accelerates training convergence and enhances exploration in the state space, outperforming current approaches on standard benchmark tasks.

This work addresses a fundamental limitation in conventional Bayesian experimental design, which relies on prior-to-posterior uncertainty reduction and yields an intractable objective that is doubly hard to evaluate and poorly aligned with downstream tasks. By reframing the problem through decision theory, the authors formulate it as optimizing the expected future loss (EFL) of downstream actions, thereby reducing the objective to a singly intractable form that obviates explicit posterior or marginal likelihood computation. They introduce a stochastic gradient method that jointly optimizes both the experimental design and the action policy, requiring only samples from the joint parameter–data model and evaluations of the loss function. This approach naturally accommodates implicit modeling and task-specific customization, demonstrating marked improvements over existing methods in both optimization efficiency and task adaptability.