The alignment of large language models (LLMs) is crucial for generating helpful and harmless content. Existing approaches leverage preference-based human feedback data to learn the reward function and align the LLM with the feedback data. However, these approaches focus on modeling the reward difference between the chosen and rejected demonstrations, rather than directly modeling the true reward from each demonstration. Moreover, these approaches assume that the reward is only obtained at the end of the sentence, which overlooks the modeling of intermediate rewards. These issues lead to insufficient use of training signals in the feedback data, limiting the representation and generalization ability of the reward and potentially resulting in reward hacking. In this paper, we formulate LLM alignment as a Bayesian Inverse Reinforcement Learning (BIRL) problem and propose a novel training objective, Approximated Variational Alignment (AVA), to perform LLM alignment through Approximated Variational Reward Imitation Learning (AVRIL). The BIRL formulation facilitates intermediate reward modeling and direct reward modeling on each single demonstration, which enhances the utilization of training signals in the feedback data. Experiments show that AVA outperforms existing LLM alignment approaches in reward modeling, RL fine-tuning, and direct optimization.
Inverse reinforcement learning (IRL) is the problem of inferring a reward function from expert behavior. There are several approaches to IRL, but most are designed to learn a Markovian reward. However, a reward function might be non-Markovian, depending on more than just the current state, such as a reward machine (RM). Although there has been recent work on inferring RMs, it assumes access to the reward signal, absent in IRL. We propose a Bayesian IRL (BIRL) framework for inferring RMs directly from expert behavior, requiring significant changes to the standard framework. We define a new reward space, adapt the expert demonstration to include history, show how to compute the reward posterior, and propose a novel modification to simulated annealing to maximize this posterior. We demonstrate that our method performs well when optimizing according to its inferred reward and compares favorably to an existing method that learns exclusively binary non-Markovian rewards.
Inverse reinforcement learning (IRL) methods infer an agent's reward function using demonstrations of expert behavior. A Bayesian IRL approach models a distribution over candidate reward functions, capturing a degree of uncertainty in the inferred reward function. This is critical in some applications, such as those involving clinical data. Typically, Bayesian IRL algorithms require large demonstration datasets, which may not be available in practice. In this work, we incorporate existing domain-specific data to achieve better posterior concentration rates. We study a common setting in clinical and biological applications where we have access to expert demonstrations and known reward functions for a set of training tasks. Our aim is to learn the reward function of a new test task given limited expert demonstrations. Existing Bayesian IRL methods impose restrictions on the form of input data, thus limiting the incorporation of training task data. To better leverage information from training tasks, we introduce kernel density Bayesian inverse reinforcement learning (KD-BIRL). Our approach employs a conditional kernel density estimator, which uses the known reward functions of the training tasks to improve the likelihood estimation across a range of reward functions and demonstration samples. Our empirical results highlight KD-BIRL's faster concentration rate in comparison to baselines, particularly in low test task expert demonstration data regimes. Additionally, we are the first to provide theoretical guarantees of posterior concentration for a Bayesian IRL algorithm. Taken together, this work introduces a principled and theoretically grounded framework that enables Bayesian IRL to be applied across a variety of domains.
This review introduces a novel deformable image registration paradigm that exploits Markov random field formulation and powerful discrete optimization algorithms. We express deformable registration as a minimal cost graph problem, where nodes correspond to the deformation grid, a node's connectivity corresponds to regularization constraints, and labels correspond to 3D deformations. To cope with both iconic and geometric (landmark-based) registration, we introduce two graphical models, one for each subproblem. The two graphs share interconnected variables, leading to a modular, powerful, and flexible formulation that can account for arbitrary image-matching criteria, various local deformation models, and regularization constraints. To cope with the corresponding optimization problem, we adopt two optimization strategies: a computationally efficient one and a tight relaxation alternative. Promising results demonstrate the potential of this approach. Discrete methods are an important new trend in medical image registration, as they provide several improvements over the more traditional continuous methods. This is illustrated with several key examples where the presented framework outperforms existing general-purpose registration methods in terms of both performance and computational complexity. Our methods become of particular interest in applications where computation time is a critical issue, as in intraoperative imaging, or where the huge variation in data demands complex and application-specific matching criteria, as in large-scale multimodal population studies. The proposed registration framework, along with a graphical interface and corresponding publications, is available for download for research purposes (for Windows and Linux platforms) from http://www.mrf-registration.net.
Image registration is a common task for many biomedical analysis applications. The present work focuses on the benchmarking of registration methods on differently stained histological slides. This is a challenging task due to the differences in the appearance model, the repetitive texture of the details and the large image size, between other issues. Our benchmarking data is composed of 616 image pairs at two different scales — average image diagonal 2.4k and 5k pixels. We compare eleven fully automatic registration methods covering the widely used similarity measures. For each method, the best parameter configuration is found and subsequently applied to all the image pairs. The performance of the algorithms is evaluated from several perspectives — the registrations (in)accuracy on manually annotated landmarks, the method robustness and its computation time.
This report presents a generic image registration benchmark with automatic evaluation using landmark annotations. The key features of the BIRL framework are: easily extendable, performance evaluation, parallel experimentation, simple visualisations, experiment's time-out limit, resuming unfinished experiments. From the research practice, we identified and focused on these two main use-cases: (a) comparison of user's (newly developed) method with some State-of-the-Art (SOTA) methods on a common dataset and (b) experimenting SOTA methods on user's custom dataset (which should contain landmark annotation). Moreover, we present an integration of several standard image registration methods aiming at biomedical imaging into the BIRL framework. This report also contains experimental results of these SOTA methods on the CIMA dataset, which is a dataset of Whole Slice Imaging (WSI) from histology/pathology containing several multi-stain tissue samples from three tissue kinds. Source and results: https://borda.github.io/BIRL
One of the most challenging problems in modern neuroimaging is detailed characterization of neurodegeneration. Quantifying spatial and longitudinal atrophy patterns is an important component of this process. These spatiotemporal signals will aid in discriminating between related diseases, such as frontotemporal dementia (FTD) and Alzheimer's disease (AD), which manifest themselves in the same at-risk population. Here, we develop a novel symmetric image normalization method (SyN) for maximizing the cross-correlation within the space of diffeomorphic maps and provide the Euler-Lagrange equations necessary for this optimization. We then turn to a careful evaluation of our method. Our evaluation uses gold standard, human cortical segmentation to contrast SyN's performance with a related elastic method and with the standard ITK implementation of Thirion's Demons algorithm. The new method compares favorably with both approaches, in particular when the distance between the template brain and the target brain is large. We then report the correlation of volumes gained by algorithmic cortical labelings of FTD and control subjects with those gained by the manual rater. This comparison shows that, of the three methods tested, SyN's volume measurements are the most strongly correlated with volume measurements gained by expert labeling. This study indicates that SyN, with cross-correlation, is a reliable method for normalizing and making anatomical measurements in volumetric MRI of patients and at-risk elderly individuals.
We introduce a new generative model for human planning under the Bayesian Inverse Reinforcement Learning (BIRL) framework which takes into account the fact that humans often plan using hierarchical strategies. We describe the Bayesian Inverse Hierarchical RL (BIHRL) algorithm for inferring the values of hierarchical planners, and use an illustrative toy model to show that BIHRL retains accuracy where standard BIRL fails. Furthermore, BIHRL is able to accurately predict the goals of `Wikispeedia' game players, with inclusion of hierarchical structure in the model resulting in a large boost in accuracy. We show that BIHRL is able to significantly outperform BIRL even when we only have a weak prior on the hierarchical structure of the plans available to the agent, and discuss the significant challenges that remain for scaling up this framework to more realistic settings.