Gliomas are the most common primary brain malignancies, with different degrees of aggressiveness, variable prognosis and various heterogeneous histologic sub-regions, i.e., peritumoral edematous/invaded tissue, necrotic core, active and non-enhancing core. This intrinsic heterogeneity is also portrayed in their radio-phenotype, as their sub-regions are depicted by varying intensity profiles disseminated across multi-parametric magnetic resonance imaging (mpMRI) scans, reflecting varying biological properties. Their heterogeneous shape, extent, and location are some of the factors that make these tumors difficult to resect, and in some cases inoperable. The amount of resected tumor is a factor also considered in longitudinal scans, when evaluating the apparent tumor for potential diagnosis of progression. Furthermore, there is mounting evidence that accurate segmentation of the various tumor sub-regions can offer the basis for quantitative image analysis towards prediction of patient overall survival. This study assesses the state-of-the-art machine learning (ML) methods used for brain tumor image analysis in mpMRI scans, during the last seven instances of the International Brain Tumor Segmentation (BraTS) challenge, i.e., 2012-2018. Specifically, we focus on i) evaluating segmentations of the various glioma sub-regions in pre-operative mpMRI scans, ii) assessing potential tumor progression by virtue of longitudinal growth of tumor sub-regions, beyond use of the RECIST/RANO criteria, and iii) predicting the overall survival from pre-operative mpMRI scans of patients that underwent gross total resection. Finally, we investigate the challenge of identifying the best ML algorithms for each of these tasks, considering that apart from being diverse on each instance of the challenge, the multi-institutional mpMRI BraTS dataset has also been a continuously evolving/growing dataset.

We present a new approach to solve the 2+1 Teukolsky equation for gravitational perturbations of a Kerr black hole. Our approach relies on a new horizon penetrating, hyperboloidal foliation of Kerr spacetime and spatial compactification. In particular, we present a framework for waveform generation from point-particle perturbations. Extensive tests of a time domain implementation in the code {\it Teukode} are presented. The code can efficiently deliver waveforms at future null infinity. As a first application of the method, we compute the gravitational waveforms from inspiraling and coalescing black-hole binaries in the large-mass-ratio limit. The smaller mass black hole is modeled as a point particle whose dynamics is driven by an effective-one-body-resummed analytical radiation reaction force. We compare the analytical angular momentum loss to the gravitational wave angular momentum flux. We find that higher-order post-Newtonian corrections are needed to improve the consistency for rapidly spinning binaries. Close to merger, the subdominant multipolar amplitudes (notably the $m=0$ ones) are enhanced for retrograde orbits with respect to prograde ones. We argue that this effect mirrors nonnegligible deviations from circularity of the dynamics during the late-plunge and merger phase. We compute the gravitational wave energy flux flowing into the black hole during the inspiral using a time-domain formalism proposed by Poisson. Finally, a self-consistent, iterative method to compute the gravitational wave fluxes at leading-order in the mass of the particle is presented. For a specific case study with $\hat{a}$=0.9, a simulation that uses the consistent flux differs from one that uses the analytical flux by $\sim35$ gravitational wave cycles over a total of about $250$ cycles. In this case the horizon absorption accounts for about $+5$ gravitational wave cycles.

Large language models (LLMs) are revolutionizing self driving laboratories (SDLs) for materials research, promising unprecedented acceleration of scientific discovery. However, current SDL implementations rely on rigid protocols that fail to capture the adaptability and intuition of expert scientists in dynamic experimental settings. We introduce Artificially Intelligent Lab Assistant (AILA), a framework automating atomic force microscopy through LLM driven agents. Further, we develop AFMBench a comprehensive evaluation suite challenging AI agents across the complete scientific workflow from experimental design to results analysis. We find that state of the art models struggle with basic tasks and coordination scenarios. Notably, Claude 3.5 sonnet performs unexpectedly poorly despite excelling in materials domain question answering (QA) benchmarks, revealing that domain specific QA proficiency does not necessarily translate to effective agentic capabilities. Additionally, we observe that LLMs can deviate from instructions, raising safety alignment concerns for SDL applications. Our ablations reveal that multi agent frameworks outperform single-agent architectures. We also observe significant prompt fragility, where slight modifications in prompt structure cause substantial performance variations in capable models like GPT 4o. Finally, we evaluate AILA's effectiveness in increasingly advanced experiments AFM calibration, feature detection, mechanical property measurement, graphene layer counting, and indenter detection. Our findings underscore the necessity for rigorous benchmarking protocols and prompt engineering strategies before deploying AI laboratory assistants in scientific research environments.

Objective Structured Clinical Examinations (OSCEs) are essential for medical training, but they require significant resources, including professional actors and expert medical feedback. Although Large Language Models (LLMs) have introduced text-based virtual patients for communication practice, these simulations often lack the capability for richer, non-textual interactions. This paper presents a novel framework that significantly enhances LLM-based simulated patients by equipping them with action spaces, thereby enabling more realistic and dynamic patient behaviors that extend beyond text. Furthermore, our system incorporates virtual tutors that provide students with instant, personalized feedback on their performance at any time during these simulated encounters. We have conducted a rigorous evaluation of the framework's real-time performance, including system latency and component accuracy. Preliminary evaluations with medical experts assessed the naturalness and coherence of the simulated patients, as well as the usefulness and appropriateness of the virtual tutor's assessments. This innovative system provides medical students with a low-cost, accessible platform for personalized OSCE preparation at home.

We investigate the generic behaviour of marginally trapped tubes (roughly time-evolved apparent horizons) using simple, spherically symmetric examples of dust and scalar field collapse/accretion onto pre-existing black holes. We find that given appropriate physical conditions the evolution of the marginally trapped tube may be either null, timelike, or spacelike and further that the marginally trapped two-sphere cross-sections may either expand or contract in area. Spacelike expansions occur when the matter falling into a black hole satisfies $\rho - P \leq 1/A$, where $A$ is the area of the horizon while $\rho$ and $P$ are respectively the density and pressure of the matter. Timelike evolutions occur when $(\rho - P)$ is greater than this cut-off and so would be expected to be more common for large black holes. Physically they correspond to horizon "jumps" as extreme conditions force the formation of new horizons outside of the old.

Euclid is expected to establish new state-of-the-art constraints on extensions beyond the standard LCDM cosmological model by measuring the positions and shapes of billions of galaxies. Specifically, its goal is to shed light on the nature of dark matter and dark energy. Achieving this requires developing and validating advanced statistical tools and theoretical prediction software capable of testing extensions of the LCDM model. In this work, we describe how the Euclid likelihood pipeline, Cosmology Likelihood for Observables in Euclid (CLOE), has been extended to accommodate alternative cosmological models and to refine the theoretical modelling of Euclid primary probes. In particular, we detail modifications made to CLOE to incorporate the magnification bias term into the spectroscopic two-point correlation function of galaxy clustering. Additionally, we explain the adaptations made to CLOE's implementation of Euclid primary photometric probes to account for massive neutrinos and modified gravity extensions. Finally, we present the validation of these CLOE modifications through dedicated forecasts on synthetic Euclid-like data by sampling the full posterior distribution and comparing with the results of previous literature. In conclusion, we have identified in this work several functionalities with regards to beyond-LCDM modelling that could be further improved within CLOE, and outline potential research directions to enhance pipeline efficiency and flexibility through novel inference and machine learning techniques.

These lectures aim to provide a basic introduction to dispersive methods and their modern applications to the phenomenology of the Standard Model at low energy. This approach exploits analyticity properties of Green functions and scattering amplitude, often combined with unitarity constraints. To find a logically coherent set of topics in this vast subject, I start with the two-point Green's function, show that this needs the three-point function as input which in turn needs the four-point function. The sequence stops here, just like these lectures, because the four-point function is related only to itself (if one ignores inelastic effects), I will discuss these dispersion relations both in the case of toy models, simple scalar theories, as well as in the phenomenologically relevant case of QCD.

We solve the generalised quantum Stein's lemma, proving that the Stein exponent associated with entanglement testing, namely, the quantum hypothesis testing task of distinguishing between $n$ copies of an entangled state $\rho_{AB}$ and a generic separable state $\sigma_{A^n:B^n}$, equals the regularised relative entropy of entanglement. Not only does this determine the ultimate performance of entanglement testing, but it also establishes the reversibility of all quantum resource theories under asymptotically resource non-generating operations, with the regularised relative entropy of resource governing the asymptotic transformation rate between any two quantum states. As a by-product, we prove that the same Stein exponent can also be achieved when the null hypothesis is only approximately i.i.d., in the sense that it can be modelled by an 'almost power state'. To solve the problem we introduce two techniques. The first is a procedure that we call 'blurring', which, informally, transforms a permutationally symmetric state by making it more evenly spread across nearby type classes. Blurring alone suffices to prove the generalised Stein's lemma in the fully classical case, but not in the quantum case. Our second technical innovation, therefore, is to perform a second quantisation step to lift the problem to an infinite-dimensional bosonic quantum system; we then solve it there by using techniques from continuous-variable quantum information. Rather remarkably, the second-quantised action of the blurring map corresponds to a pure loss channel. A careful examination of this second quantisation step is the core of our quantum solution.

We investigate the physics of a small group of quantum states defined above the sharply defined ground state of a chaotic ensemble. This `universality class of the first levels' (UFL) is realized in the majority of `synthetic' random matrix models but, for all we know, in only one microscopically defined system: low-dimensional gravity. We discuss the physical properties of these states, notably their exceptional rigidity against external perturbations, as quantified by the so-called quantum state fidelity. Examining these structures through the lenses of random matrix and string theory, we highlight their relevance to the physics of low-dimensional holographic principles.

Quantum signal processing (QSP) relies on a historically costly pre-processing step, "QSP-processing/phase-factor finding." QSP-processing is now a developed topic within quantum algorithms literature, and a beginner accessible review of QSP-processing is overdue. This work provides a whirlwind tour through QSP conventions and pre-processing methods, beginning from a pedagogically accessible QSP convention. We then review QSP conventions associated with three common polynomial types: real polynomials with definite parity, sums of reciprocal/anti-reciprocal Chebyshev polynomials, and complex polynomials. We demonstrate how the conventions perform with respect to three criteria: circuit length, polynomial conditions, and pre-processing methods. We then review the recently introduced Wilson method for QSP-processing and give conditions where it can succeed with bound error. Although the resulting bound is not computationally efficient, we demonstrate that the method succeeds with linear error propagation for relevant target polynomials and precision regimes, including the Jacobi-Anger expansion used in Hamiltonian simulation algorithms. We then apply our benchmarks to three QSP-processing methods for QSP circuits and show that a method introduced by Berntson and Sünderhauf outperforms both the Wilson method and the standard optimization strategy for complex polynomials.

We discuss an enhancement of the Brown-Henneaux boundary conditions in three-dimensional AdS General Relativity to encompass Weyl transformations of the boundary metric. The resulting asymptotic symmetry algebra, after a field-dependent redefinition of the generators, is a direct sum of two copies of the Witt algebra and the Weyl abelian sector. The charges associated to Weyl transformations are non-vanishing, integrable but not conserved due to a flux driven by the Weyl anomaly coefficient. The charge algebra admits an additional non-trivial central extension in the Weyl sector, related to the well-known Weyl anomaly. We then construct the holographic Weyl current and show that it satisfies an anomalous Ward-Takahashi identity of the boundary theory.

Thermodynamics teaches that if a system initially off-equilibrium is coupled to work sources, the maximum work that it may yield is governed by its energy and entropy. For finite systems this bound is usually not reachable. The maximum extractable work compatible with quantum mechanics (``ergotropy'') is derived and expressed in terms of the density matrix and the Hamiltonian. It is related to the property of majorization: more major states can provide more work. Scenarios of work extraction that contrast the thermodynamic intuition are discussed, e.g. a state with larger entropy than another may produce more work, while correlations may increase or reduce the ergotropy.

We test the gauge/gravity duality between the matrix model and type IIA string theory at low temperatures with unprecedented accuracy. To this end, we perform lattice Monte Carlo simulations of the Berenstein-Maldacena-Nastase (BMN) matrix model, which is the one-parameter deformation of the Banks-Fischler-Shenker-Susskind (BFSS) matrix model, taking both the large $N$ and continuum limits. We leverage the fact that sufficiently small flux parameters in the BMN matrix model have a negligible impact on the energy of the system while stabilizing the flat directions so that simulations at smaller $N$ than in the BFSS matrix model are possible. Hence, we can perform a precision measurement of the large $N$ continuum energy at the lowest temperatures to date. The energy is in perfect agreement with supergravity predictions including estimations of $\alpha'$-corrections from previous simulations. At the lowest temperature where we can simulate efficiently ($T=0.25\lambda^{1/3}$, where $\lambda$ is the 't Hooft coupling), the difference in energy to the pure supergravity prediction is less than $10\%$. Furthermore, we can extract the coefficient of the $1/N^4$ corrections at a fixed temperature with good accuracy, which was previously unknown.

Existing language and vision models achieve impressive performance in image-text understanding. Yet, it is an open question to what extent they can be used for language understanding in 3D environments and whether they implicitly acquire 3D object knowledge, e.g. about different views of an object. In this paper, we investigate whether a state-of-the-art language and vision model, CLIP, is able to ground perspective descriptions of a 3D object and identify canonical views of common objects based on text queries. We present an evaluation framework that uses a circling camera around a 3D object to generate images from different viewpoints and evaluate them in terms of their similarity to natural language descriptions. We find that a pre-trained CLIP model performs poorly on most canonical views and that fine-tuning using hard negative sampling and random contrasting yields good results even under conditions with little available training data.

We propose an orbifold lattice formulation of QCD suitable for quantum simulations. We show explicitly how to encode gauge degrees of freedom into qubits using noncompact variables, and how to write down a simple truncated Hamiltonian in the coordinate basis. We show that SU(3) gauge group variables and quarks in the fundamental representation can be implemented straightforwardly on qubits, for arbitrary truncation of the gauge manifold.

One of the main problems for the future of practical quantum computing is to stabilize the computation against unwanted interactions with the environment and imperfections in the applied operations. Existing proposals for quantum memories and quantum channels require gates with asymptotically zero error to store or transmit an input quantum state for arbitrarily long times or distances with fixed error. In this report a method is given which has the property that to store or transmit a qubit with maximum error $\epsilon$ requires gates with error at most $c\epsilon$ and storage or channel elements with error at most $\epsilon$, independent of how long we wish to store the state or how far we wish to transmit it. The method relies on using concatenated quantum codes with hierarchically implemented recovery operations. The overhead of the method is polynomial in the time of storage or the distance of the transmission. Rigorous and heuristic lower bounds for the constant $c$ are given.

We compute and analyze the gravitational waveform emitted to future null infinity by a system of two black holes in the large mass ratio limit. We consider the transition from the quasi-adiabatic inspiral to plunge, merger, and ringdown. The relative dynamics is driven by a leading order in the mass ratio, 5PN-resummed, effective-one-body (EOB), analytic radiation reaction. To compute the waveforms we solve the Regge-Wheeler-Zerilli equations in the time-domain on a spacelike foliation which coincides with the standard Schwarzschild foliation in the region including the motion of the small black hole, and is globally hyperboloidal, allowing us to include future null infinity in the computational domain by compactification. This method is called the hyperboloidal layer method, and is discussed here for the first time in a study of the gravitational radiation emitted by black hole binaries. We consider binaries characterized by five mass ratios, $\nu=10^{-2,-3,-4,-5,-6}$, that are primary targets of space-based or third-generation gravitational wave detectors. We show significative phase differences between finite-radius and null-infinity waveforms. We test, in our context, the reliability of the extrapolation procedure routinely applied to numerical relativity waveforms. We present an updated calculation of the gravitational recoil imparted to the merger remnant by the gravitational wave emission. As a self consistency test of the method, we show an excellent fractional agreement (even during the plunge) between the 5PN EOB-resummed mechanical angular momentum loss and the gravitational wave angular momentum flux computed at null infinity. New results concerning the radiation emitted from unstable circular orbits are also presented.

Decision trees are a classic model for summarizing and classifying data. To enhance interpretability and generalization properties, it has been proposed to favor small decision trees. Accordingly, in the minimum-size decision tree training problem (MSDT), the input is a set of training examples in $\mathbb{R}^d$ with class labels and we aim to find a decision tree that classifies all training examples correctly and has a minimum number of nodes. MSDT is NP-hard and therefore presumably not solvable in polynomial time. Nevertheless, Komusiewicz et al. [ICML '23] developed a promising algorithmic paradigm called witness trees which solves MSDT efficiently if the solution tree is small. In this work, we test this paradigm empirically. We provide an implementation, augment it with extensive heuristic improvements, and scrutinize it on standard benchmark instances. The augmentations achieve a mean 324-fold (median 84-fold) speedup over the naive implementation. Compared to the state of the art they achieve a mean 32-fold (median 7-fold) speedup over the dynamic programming based MurTree solver [Demirović et al., J. Mach. Learn. Res. '22] and a mean 61-fold (median 25-fold) speedup over SAT-based implementations [Janota and Morgado, SAT '20]. As a theoretical result we obtain an improved worst-case running-time bound for MSDT.

Detecting vehicles and representing their position and orientation in the three dimensional space is a key technology for autonomous driving. Recently, methods for 3D vehicle detection solely based on monocular RGB images gained popularity. In order to facilitate this task as well as to compare and drive state-of-the-art methods, several new datasets and benchmarks have been published. Ground truth annotations of vehicles are usually obtained using lidar point clouds, which often induces errors due to imperfect calibration or synchronization between both sensors. To this end, we propose Cityscapes 3D, extending the original Cityscapes dataset with 3D bounding box annotations for all types of vehicles. In contrast to existing datasets, our 3D annotations were labeled using stereo RGB images only and capture all nine degrees of freedom. This leads to a pixel-accurate reprojection in the RGB image and a higher range of annotations compared to lidar-based approaches. In order to ease multitask learning, we provide a pairing of 2D instance segments with 3D bounding boxes. In addition, we complement the Cityscapes benchmark suite with 3D vehicle detection based on the new annotations as well as metrics presented in this work. Dataset and benchmark are available online.