Any well-behaved generative model over a variable $\mathbf{x}$ can be expressed as a deterministic transformation of an exogenous (outsourced’) Gaussian noise variable $\mathbf{z}$: $\mathbf{x}=f_\theta(\mathbf{z})$. In such a model (eg, a VAE, GAN, or continuous-time flow-based model), sampling of the target variable $\mathbf{x} \sim p_\theta(\mathbf{x})$ is straightforward, but sampling from a posterior distribution of the form $p(\mathbf{x}\mid\mathbf{y}) \propto p_\theta(\mathbf{x})r(\mathbf{x},\mathbf{y})$, where $r$ is a constraint function depending on an auxiliary variable $\mathbf{y}$, is generally intractable. We propose to amortize the cost of sampling from such posterior distributions with diffusion models that sample a distribution in the noise space ($\mathbf{z}$). These diffusion samplers are trained by reinforcement learning algorithms to enforce that the transformed samples $f_\theta(\mathbf{z})$ are distributed according to the posterior in the data space ($\mathbf{x}$). For many models and constraints, the posterior in noise space is smoother than in data space, making it more suitable for amortized inference. Our method enables conditional sampling under unconditional GAN, (H)VAE, and flow-based priors, comparing favorably with other inference methods. We demonstrate the proposed outsourced diffusion sampling in several experiments with large pretrained prior models: conditional image generation, reinforcement learning with human feedback, and protein structure generation.

As generative AI systems become competent and democratized in science, business, and government, deeper insight into their failure modes now poses an acute need. The occasional volatility in their behavior, such as the propensity of transformer models to hallucinate, impedes trust and adoption of emerging AI solutions in high-stakes areas. In the present work, we establish how and when hallucinations arise in pre-trained transformer models through concept representations captured by sparse autoencoders, under scenarios with experimentally controlled uncertainty in the input space. Our systematic experiments reveal that the number of semantic concepts used by the transformer model grows as the input information becomes increasingly unstructured. In the face of growing uncertainty in the input space, the transformer model becomes prone to activate coherent yet input-insensitive semantic features, leading to hallucinated output. At its extreme, for pure-noise inputs, we identify a wide variety of robustly triggered and meaningful concepts in the intermediate activations of pre-trained transformer models, whose functional integrity we confirm through targeted steering. We also show that hallucinations in the output of a transformer model can be reliably predicted from the concept patterns embedded in transformer layer activations. This collection of insights on transformer internal processing mechanics has immediate consequences for aligning AI models with human values, AI safety, opening the attack surface for potential adversarial attacks, and providing a basis for automatic quantification of a model’s hallucination risk.

Teleoperated medical robots have the potential to revolutionize healthcare. However, when developing systems for tasks like remote palpation, state-of-the-art literature still uses test phantoms of oversimplified geometries, due to the complexity of the required mechanical robot–patient interaction. In reality, human bodies have complex 3-D shapes and require fine-tuning of all six manipulator's degrees of freedom, controlled by the user. In this article, we argue that the implementation of depth-vision-driven autonomous dimensionality-reduction (DVD ADR) shared control can greatly improve the users' performance. The proposed control method keeps the user in control of the end-effector’s position, while automatically adjusting its orientation in order to maintain the tactile sensor normal to the phantom's surface. A depth camera and a computer vision algorithm are used to infer the phantom's shape and achieve DVD ADR shared control. Experimental results showcase how this leads to statistically significant performance improvement. Not only were the participants able to achieve more precise palpations, with up to 29.5% and 22.4% more accuracy in position and orientation, respectively, but the DVD ADR shared control allowed them to achieve a 8.8% better detection accuracy while needing 13.8% less time. The abovementioned results are all tested for statistical significance and achieved ap-value lower than 0.05.

We study the problem of training diffusion models to sample from a distribution with a given unnormalized density or energy function. We benchmark several diffusion-structured inference methods, including simulation-based variational approaches and off-policy methods (continuous generative flow networks). Our results shed light on the relative advantages of existing algorithms while bringing into question some claims from past work. We also propose a novel exploration strategy for off-policy methods, based on local search in the target space with the use of a replay buffer, and show that it improves the quality of samples on a variety of target distributions. Our code for the sampling methods and benchmarks studied is made public at [this link](https://github.com/GFNOrg/gfn-diffusion) as a base for future work on diffusion models for amortized inference.

Diffusion models have emerged as effective distribution estimators in vision, language, and reinforcement learning, but their use as priors in downstream tasks poses an intractable posterior inference problem. This paper studies *amortized* sampling of the posterior over data, $\mathbf{x}\sim p^{\rm post}(\mathbf{x})\propto p(\mathbf{x})r(\mathbf{x})$, in a model that consists of a diffusion generative model prior $p(\mathbf{x})$ and a black-box constraint or likelihood function $r(\mathbf{x})$. We state and prove the asymptotic correctness of a data-free learning objective, *relative trajectory balance*, for training a diffusion model that samples from this posterior, a problem that existing methods solve only approximately or in restricted cases. Relative trajectory balance arises from the generative flow network perspective on diffusion models, which allows the use of deep reinforcement learning techniques to improve mode coverage. Experiments illustrate the broad potential of unbiased inference of arbitrary posteriors under diffusion priors: in vision (classifier guidance), language (infilling under a discrete diffusion LLM), and multimodal data (text-to-image generation). Beyond generative modeling, we apply relative trajectory balance to the problem of continuous control with a score-based behavior prior, achieving state-of-the-art results on benchmarks in offline reinforcement learning. Code is available at [this link](https://github.com/GFNOrg/diffusion-finetuning).

Effective control and prediction of dynamical systems require appropriate handling of continuous-time and discrete, event-triggered processes. Stochastic hybrid systems (SHSs), common across engineering domains, provide a formalism for dynamical systems subject to discrete, possibly stochastic, state jumps and multi-modal continuous-time flows. Despite the versatility and importance of SHSs across applications, a general procedure for the explicit learning of both discrete events and multi-mode continuous dynamics remains an open problem. This work introduces Neural Hybrid Automata (NHAs), a recipe for learning SHS dynamics without a priori knowledge on the number, mode parameters, and inter-modal transition dynamics. NHAs provide a systematic inference method based on normalizing flows, neural differential equations, and self-supervision. We showcase NHAs on several tasks, including mode recovery and flow learning in systems with stochastic transitions, and end-to-end learning of hierarchical robot controllers.

Robotic phantoms enable advanced physical examination training before using human patients. In this article, we present an abdominal phantom for palpation training with controllable stiffness liver nodules that can also sense palpation forces. The coupled sensing and actuation approach is achieved by pneumatic control of positive-granular jammed nodules for tunable stiffness. Soft sensing is done using the variation of internal pressure of the nodules under external forces. This article makes original contributions to extend the linear region of the neo-Hookean characteristic of the mechanical behavior of the nodules by 140% compared to no-jamming conditions and to propose a method using the organ level controllable nodules as sensors to estimate palpation position and force with a root-mean-square error of 4% and 6.5%, respectively. Compared to conventional soft sensors, the method allows the phantom to sense with no interference to the simulated physiological conditions when providing quantified feedback to trainees, and to enable training following current bare-hand examination protocols without the need to wear data gloves to collect data.

To match the ever increasing standards of fresh products, and the need to reduce waste, we devise an alternative to the destructive and highly variable fruit ripeness estimation by a penetrometer. We propose a fully automatic method to assess the ripeness of mango which is non-destructive, allows the user to test multiple surface areas with a single touch and is capable of dissociating between ripe and non-ripe fruits. A custom-made gripper equipped with a capacitive tactile sensor array is used to palpate the fruit. The ripeness is estimated as mango stiffness extracted through a simplified spring model. We test the framework on a set of 25 mangoes of the Keitt variety, and compare the results to penetrometer measurements. We show it is possible to correctly classify 88% of the mango without removing the skin of the fruit. The method can be a valuable substitute for non-destructive fruit ripeness testing. To the authors knowledge, this is the first robotics ripeness estimation system based on capacitive tactile sensing technology.