Background: There is great interest in agentic LLMs, large language models that act as agents. Objectives: We review the growing body of work in this area and provide a research agenda. Methods: Agentic LLMs are LLMs that (1) reason, (2) act, and (3) interact. We organize the literature according to these three categories. Results: The research in the first category focuses on reasoning, reflection, and retrieval, aiming to improve decision making; the second category focuses on action models, robots, and tools, aiming for agents that act as useful assistants; the third category focuses on multi-agent systems, aiming for collaborative task solving and simulating interaction to study emergent social behavior. We find that works mutually benefit from results in other categories: retrieval enables tool use, reflection improves multi-agent collaboration, and reasoning benefits all categories. Conclusions: We discuss applications of agentic LLMs and provide an agenda for further research. Important applications are in medical diagnosis, logistics and financial market analysis. Meanwhile, self-reflective agents playing roles and interacting with one another augment the process of scientific research itself. Further, agentic LLMs provide a solution for the problem of LLMs running out of training data: inference-time behavior generates new training states, such that LLMs can keep learning without needing ever larger datasets. We note that there is risk associated with LLM assistants taking action in the real world—safety, liability and security are open problems—while agentic LLMs are also likely to benefit society.

Recent years have witnessed phenomenal growth in the application, and capabilities of graphical processing units (GPUs) due to their high parallel computation power at relatively low cost. However, writing a computationally efficient GPU program (kernel) is challenging and, generally, only certain specific kernel configurations lead to significant increases in performance. Auto-tuning is the process of automatically optimizing software for highly efficient execution on a target hardware platform. Auto-tuning is particularly useful for GPU programming, as a single kernel requires retuning after code changes, for different input data, and for different architectures. However, the discrete and nonconvex nature of the search space creates a challenging optimization problem. In this work, we investigate which algorithm produces the fastest kernels if the time-budget for the tuning task is varied. We conduct a survey by performing experiments on 26 different kernel spaces, from nine different GPUs, for 16 different evolutionary black-box optimization algorithms. We then analyze these results and introduce a novel metric based on the PageRank centrality concept as a tool for gaining insight into the difficulty of the optimization problem. We demonstrate that our metric correlates strongly with the observed tuning performance.

Detection of unwanted (‘foreign’) objects within products is a common procedure in many branches of industry for maintaining production quality. X-ray imaging is a fast, non-invasive and widely applicable method for foreign object detection. Deep learning has recently emerged as a powerful approach for recognizing patterns in radiographs (i.e., X-ray images), enabling automated X-ray based foreign object detection. However, these methods require a large number of training examples and manual annotation of these examples is a subjective and laborious task. In this work, we propose a Computed Tomography (CT) based method for producing training data for supervised learning of foreign object detection, with minimal labor requirements. In our approach, a few representative objects are CT scanned and reconstructed in 3D. The radiographs that are acquired as part of the CT-scan data serve as input for the machine learning method. High-quality ground truth locations of the foreign objects are obtained through accurate 3D reconstructions and segmentations. Using these segmented volumes, corresponding 2D segmentations are obtained by creating virtual projections. We outline the benefits of objectively and reproducibly generating training data in this way. In addition, we show how the accuracy depends on the number of objects used for the CT reconstructions. The results show that in this workflow generally only a relatively small number of representative objects (i.e., fewer than 10) are needed to achieve adequate detection performance in an industrial setting.

Choosing a regularization parameter for tomographic reconstruction algorithms is often a cumbersome task of trial-and-error. Although several automatic and objective criteria have been proposed, each of them yields a different “optimal” value, which may or may not correspond to the actual implicit image quality metrics one would like to optimize for. Exploration of the space of regularization parameters is computationally expensive, as it requires many reconstructions to be computed. In this paper we propose an algorithmic approach for computationally efficient exploration of the regularization parameter space, based on a pixel-wise interpolation scheme. Once a relatively small number of reconstructions have been computed for a sparse sampling of the parameters, an approximation of the reconstructed image for other parameter values can be computed instantly, thereby allowing both manual and automated selection of the most preferable parameters based on a variety of image quality metrics. We demonstrate that for three common variational reconstruction methods, our approach results in accurate approximations of the reconstructed image and that it can be used in combination with existing approaches for choosing optimal regularization parameters.

Computed Tomography (CT) is an imaging technique that allows to reconstruct volumetric information of the analyzed objects from their projections. The most popular reconstruction technique is the Filtered Back Projection (FBP). It has the advantage of being the fastest technique available, but also the disadvantage to require a high number of projections to retrieve good quality reconstructions. In this article we propose a segmentation method for tomographic volumes composed of few materials. Our method combines existing high-quality variational segmentation frameworks with the data consistency approach used in tomography and discrete tomography. We show that our algorithm performs well under high noise level and with moderately low number of projections, and that the data consistency significantly improves the segmentation, at the cost of only one FBP reconstruction and forward projection.

Diffraction contrast tomography (DCT) is an X-ray full-field imaging technique that allows for the non-destructive three-dimensional investigation of polycrystalline materials and the determination of the physical and morphological properties of their crystallographic domains, called grains. This t ask is considered more and more challenging with the increasing intra-granular deformation, also known as orientation-spread. The recent introduction of a six-dimensional reconstruction framework in DCT (6D-DCT) has proven to be able to address the intra-granular crystal orientation for moderately deformed materials. The approach used in 6D-DCT, which is an extended sampling of the six-dimensional combined position-orientation space, has a linear scaling between the number of sampled orientations, which determine the orientation-space resolution of the problem, and computer memory usage. As a result, the reconstruction of more deformed materials is limited by their high resource requirements from a memory and computational point of view, which can easily become too demanding for the currently available computer technologies. In this article we propose a super-sampling method for the orientation-space representation of the six-dimensional DCT framework that enables the reconstruction of more deformed cases by reducing the impact on system memory, at the expense of longer reconstruction times. The use of super-sampling can further improve the quality and accuracy of the reconstructions, especially in cases where memory restrictions force us to adapt to inadequate (undersampled) orientation-space sampling.

In this paper, we present a novel iterative reconstruction algorithm for discrete tomography (DT) named total variation regularized discrete algebraic reconstruction technique (TVR-DART) with automated gray value estimation. This algorithm is more robust and automated than the original DART algorithm, and is aimed at imaging of objects consisting of only a few different material compositions, each corresponding to a different gray value in the reconstruction. By exploiting two types of prior knowledge of the scanned object simultaneously, TVR-DART solves the discrete reconstruction problem within an optimization framework inspired by compressive sensing to steer the current reconstruction toward a solution with the specified number of discrete gray values. The gray values and the thresholds are estimated as the reconstruction improves through iterations. Extensive experiments from simulated data, experimental μCT, and electron tomography data sets show that TVR-DART is capable of providing more accurate reconstruction than existing algorithms under noisy conditions from a small number of projection images and/or from a small angular range. Furthermore, the new algorithm requires less effort on parameter tuning compared with the original DART algorithm. With TVR-DART, we aim to provide the tomography society with an easy-to-use and robust algorithm for DT.

X-ray computed tomography (CT) is a powerful tool for non-invasive cardiac imaging. However, radiation dose is a major issue. In this paper, we propose an iterative reconstruction method that reduces the radiation dose without compromising image quality. This is achieved by exploiting prior knowledge in two ways: the reconstructed object is assumed to consist of both stationary and dynamic regions over time and the dynamic region is assumed to have sparse structures after a proper sparsifying space-time transform. Experiments on simulation data and a real μCT cardiac mouse dataset show that, with comparable image quality, the radiation dose can be substantially reduced compared to conventional acquisition/reconstruction protocols.

The study of fluid flow through solid matter by computed tomography (CT) imaging has many applications, ranging from petroleum and aquifer engineering to biomedical, manufacturing, and environmental research. To avoid motion artifacts, current experiments are often limited to slow fluid flow dynamics. This severely limits the applicability of the technique. In this paper, a new iterative CT reconstruction algorithm for improved a temporal/spatial resolution in the imaging of fluid flow through solid matter is introduced. The proposed algorithm exploits prior knowledge in two ways. First, the time-varying object is assumed to consist of stationary (the solid matter) and dynamic regions (the fluid flow). Second, the attenuation curve of a particular voxel in the dynamic region is modeled by a piecewise constant function over time, which is in accordance with the actual advancing fluid/air boundary. Quantitative and qualitative results on different simulation experiments and a real neutron tomography data set show that, in comparison with the state-of-the-art algorithms, the proposed algorithm allows reconstruction from substantially fewer projections per rotation without image quality loss. Therefore, the temporal resolution can be substantially increased, and thus fluid flow experiments with faster dynamics can be performed.

In tomography, slight differences between the geometry of the scanner hardware and the geometric model used in the reconstruction lead to alignment artifacts. To exploit high-resolution detectors used in many applications of tomography, alignment of

Filtered Backprojection is the most widely used reconstruction method in transmission tomography. The algorithm is computationally efficient, but requires a large number of low-noise projections acquired over the full angular range to produce accurat

In computed tomography (CT), partial volume effects impede accurate segmentation of structures that are small with respect to the pixel size. In this paper, it is shown that for objects consisting of a small number of homogeneous materials, the reconstruction resolution can be substantially increased without altering the acquisition process. A super-resolution reconstruction approach is introduced that is based on discrete tomography, in which prior knowledge about the materials in the object is assumed. Discrete tomography has already been used to create reconstructions from a low number of projection angles, but in this paper, it is demonstrated that it can also be applied to increase the reconstruction resolution. Experiments on simulated and real μCT data of bone and foam structures show that the proposed method indeed leads to significantly improved structure segmentation and quantification compared with what can be achieved from conventional reconstructions.

X-ray computed tomography (CT) is a powerful tool for noninvasive imaging of time-varying objects. In the past, methods have been proposed to reconstruct images from continuously changing objects. For discretely or structurally changing objects, however, such methods fail to reconstruct high quality images, mainly because assumptions about continuity are no longer valid. In this paper, we propose a method to reconstruct structurally changing objects. Starting from the observation that there exist regions within the scanned object that remain unchanged over time, we introduce an iterative optimization routine that can automatically determine these regions and incorporate this knowledge in an algebraic reconstruction method. The proposed algorithm was validated on simulation data and experimental μCT data, illustrating its capability to reconstruct structurally changing objects more accurately in comparison to current techniques.

Filtered backprojection, one of the most widely used reconstruction methods in tomography, requires a large number of low-noise projections to yield accurate reconstructions. In many applications of tomography, complete projection data of high quality cannot be obtained, because of practical considerations. Algebraic methods tend to handle such problems better, but are computationally more expensive. In this paper, we introduce a new method that improves the filtered backprojection method by using a custom data-dependent filter that minimizes the projection error of the resulting reconstruction. We show that the computational cost of the new method is significantly lower than that of algebraic methods. Experiments on both simulation and experimental data show that the method is able to produce more accurate reconstructions than filtered backprojection based on popular static filters when presented with data with a limited number of projections or statistical noise present. Furthermore, the results show that the method produces reconstructions with similar accuracy to algebraic methods, but is faster at producing them. Finally, we show that the method can be extended to exploit certain forms of prior knowledge, improving reconstruction accuracy in specific cases.

Discrete tomography deals with tomographic reconstruction of greyscale images for which the set of possible grey levels is discrete and small. Here, we develop a discrete approximate reconstruction algorithm. Our algorithm computes an image that has only grey values belonging to a given finite set. It also guarantees that the difference between the given projections and the projections of the reconstructed discrete image is bounded. The bound, which is computable, is independent of the image size. We present reconstruction experiments for a range of phantom images and a varying number of grey values.

Discrete tomography (DT) is concerned with the tomographic reconstruction of images that consist of only a small number of grey levels. Recently, DART, a practical algorithm was introduced for discrete tomography, which was validated in the domain of X-ray computed and electron tomography. In this paper, DART is introduced for magnetic resonance imaging. Using simulation experiments, it is shown that the proposed MRI-DART algorithm is capable of computing high quality reconstructions from substantially fewer data compared to state-of-the-art MRI reconstruction methods.

Image reconstruction from a small number of projections is a challenging problem in tomography. Advanced algorithms that incorporate prior knowledge can sometimes produce accurate reconstructions, but they typically require long computation times. Furthermore, the required prior knowledge can be very specific, limiting the type of images that can be reconstructed. Here, we present a reconstruction method that automatically learns prior knowledge using an artificial neural network. We show that this method can be viewed as a combination of filtered backprojection steps, and, therefore, has a relatively low computational cost. Results for two different cases show that the new method is able to use the learned information to produce high quality reconstructions in a short time, even when presented with a small number of projections.

Computed tomography (CT) is a technique for noninvasive imaging of physical objects. In the discrete algebraic reconstruction technique (DART), prior knowledge about the material's densities is exploited to obtain high quality reconstructed images from a limited number of its projections. In practice, this prior knowledge is typically not readily available. Here, a fully automatic method, called projection distance minimization DART (PDM-DART), is proposed in which the optimal grey level parameters are adaptively estimated during the reconstruction process. To apply PDM-DART, only the number of different grey levels should be known in advance. Simulation as well as real μCT experiments show that PDM-DART is capable of computing reconstructed images of which the quality is similar to reconstructions computed by conventional DART based on exact prior knowledge, thereby eliminating the need for tedious and error-prone user interaction.

Most reconstruction algorithms for transmission tomography can be subdivided in two classes: variants of Filtered Backprojection (FBP) and iterative algebraic methods. Filtered backprojection is very fast and yields accurate results when a large number of projections are available, with high SNR and a full angular range. Algebraic methods require much more computation time, yet they are more flexible in dealing with limited data problems and noise. In this paper we propose an algorithm that combines the best of these two approaches: for a given linear algebraic method, a filter is computed that can be used within the FBP algorithm. The FBP reconstructions that result from using this filter strongly resemble the algebraic reconstructions and have many of their favorable properties, while the required reconstruction time is similar to standard-FBP. Based on a series of experiments, for both simulation data and experimental data, we demonstrate the merits of the proposed algorithm.

In this paper, we present an iterative reconstruction algorithm for discrete tomography, called discrete algebraic reconstruction technique (DART). DART can be applied if the scanned object is known to consist of only a few different compositions, each corresponding to a constant gray value in the reconstruction. Prior knowledge of the gray values for each of the compositions is exploited to steer the current reconstruction towards a reconstruction that contains only these gray values. Based on experiments with both simulated CT data and experimental μCT data, it is shown that DART is capable of computing more accurate reconstructions from a small number of projection images, or from a small angular range, than alternative methods. It is also shown that DART can deal effectively with noisy projection data and that the algorithm is robust with respect to errors in the estimation of the gray values.

In this paper, we present a novel approach to segment dense, homogeneous objects in a tomographic reconstruction (or tomogram). A popular method to extract such objects from a tomogram is global thresholding, in which the threshold value is determined from the image histogram. However, accurate threshold selection is not straightforward, since, due to noise or artefacts in the reconstruction, the histogram does not always contain a clear, separate peak for the dense object. We propose a new threshold estimation approach, segmentation inconsistency minimization, that exploits the available projection data to determine the optimal global threshold. The proposed algorithm was tested on simulation data and on experimental μCT data. The results show that this method results in more accurate segmentations, compared to alternative threshold selection methods.

Discrete tomography is concerned with the reconstruction of images that are defined on a discrete set of lattice points from their projections in several directions. The range of values that can be assigned to each lattice point is typically a small discrete set. In this paper we present a framework for studying these problems from an algebraic perspective, based on ring theory and commutative algebra. A principal advantage of this abstract setting is that a vast body of existing theory becomes accessible for solving discrete tomography problems. We provide proofs of several new results on the structure of dependencies between projections, including a discrete analogon of the well-known Helgason–Ludwig consistency conditions from continuous tomography.

Grey value thresholding is a segmentation technique commonly applied to tomographic image reconstructions. Many procedures have been proposed to optimally select the grey value thresholds based on the tomogram data only (e.g., using the image histogram). In this paper, a projection distance minimization (PDM) method is presented that uses the tomographic projection data to determine optimal thresholds. These thresholds are computed by minimizing the distance between the forward projection of the segmented image and the measured projection data. An important contribution of the current paper is the efficient implementation of the forward projection method, which makes the use of the original projection data as a segmentation criterion feasible. Simulation experiments applied to various phantom images show that our proposed method obtains superior results compared to established histogram-based methods.

Discrete tomography (DT) is concerned with the tomographic reconstruction of images that consist of only a small number of gray levels. DT reconstruction problems are usually underdetermined. Therefore, incorporation of heuristic rules to guide the reconstruction algorithm towards an optimal as well as intuitive solution would be valuable. In this paper, we introduce DART: a new, heuristic DT algorithm that is based on an iterative algebraic reconstruction method. Starting from a continuous reconstruction, a discrete image is reconstructed by consistent updating of border pixels. Using simulation experiments, it is shown that the DART algorithm is capable of computing high quality reconstructions from substantially fewer projections than required for conventional continuous tomography.