Search

We introduce a simple and versatile model that enables controlled design of rugged energy landscapes that realize different types of Parisi overlap distributions. This model captures quintessential aspects of Replica Symmetry Breaking (RSB) theory and may afford additional insights into complex systems and numerical methods for their analysis.

We introduce ``local uncertainty relations'' in thermal many-body systems, from which fundamental bounds in quantum systems can be derived. These lead to universal non-relativistic speed limits (independent of interaction range) and transport coefficient bounds (e.g., those of the diffusion constant and viscosity) that are compared against experimental data., Comment: 13 pages, 4 figures, 12 tables

Theoretically describing feature learning in neural networks is crucial for understanding their expressive power and inductive biases, motivating various approaches. Some approaches describe network behavior after training through a simple change in kernel scale from initialization, resulting in a generalization power comparable to a Gaussian process. Conversely, in other approaches training results in the adaptation of the kernel to the data, involving complex directional changes to the kernel. While these approaches capture different facets of network behavior, their relationship and respective strengths across scaling regimes remains an open question. This work presents a theoretical framework of multi-scale adaptive feature learning bridging these approaches. Using methods from statistical mechanics, we derive analytical expressions for network output statistics which are valid across scaling regimes and in the continuum between them. A systematic expansion of the network's probability distribution reveals that mean-field scaling requires only a saddle-point approximation, while standard scaling necessitates additional correction terms. Remarkably, we find across regimes that kernel adaptation can be reduced to an effective kernel rescaling when predicting the mean network output of a linear network. However, even in this case, the multi-scale adaptive approach captures directional feature learning effects, providing richer insights than what could be recovered from a rescaling of the kernel alone., Comment: 24 pages, 6 figures

We extend the Coupon Collector's Problem (CCP) and present a novel generalized model, referred as the k-LCCP problem, where one is interested in recovering a bipartite graph with a perfect matching, which represents the coupons and their matching labels. We show two extra-extensions to this variation: the heterogeneous sample size case (K-LCCP) and the partly recovering case.

Despite tremendous recent progress, generative video models still struggle to capture real-world motion, dynamics, and physics. We show that this limitation arises from the conventional pixel reconstruction objective, which biases models toward appearance fidelity at the expense of motion coherence. To address this, we introduce VideoJAM, a novel framework that instills an effective motion prior to video generators, by encouraging the model to learn a joint appearance-motion representation. VideoJAM is composed of two complementary units. During training, we extend the objective to predict both the generated pixels and their corresponding motion from a single learned representation. During inference, we introduce Inner-Guidance, a mechanism that steers the generation toward coherent motion by leveraging the model's own evolving motion prediction as a dynamic guidance signal. Notably, our framework can be applied to any video model with minimal adaptations, requiring no modifications to the training data or scaling of the model. VideoJAM achieves state-of-the-art performance in motion coherence, surpassing highly competitive proprietary models while also enhancing the perceived visual quality of the generations. These findings emphasize that appearance and motion can be complementary and, when effectively integrated, enhance both the visual quality and the coherence of video generation. Project website: https://hila-chefer.github.io/videojam-paper.github.io/

In this work, we address the problem of a resource-efficient formulation of non-Abelian LGTs by focusing on the difficulty of simulating fermionic degrees of freedom and the Hilbert space redundancy. First, we show a procedure that removes the matter and improves the efficiency of the hardware resources. We demonstrate it for the simplest non-Abelian group addressable with this procedure, $\mathbb{D}_8$, both in the cases of one (1D) and two (2D) spatial dimensions. Then, with the objective of running a variational quantum simulation on real quantum hardware, we map the $\mathbb{D}_8$ lattice gauge theory onto qudit systems with local interactions. We propose a variational scheme for the qudit system with a local Hamiltonian, which can be implemented on a universal qudit quantum device as the one developed in $\href{https://doi.org/10.1038/s41567-022-01658-0}{[Nat. Phys. 18, 1053 (2022)]}$. Our results show the effectiveness of the matter-removing procedure, solving the redundancy problem and reducing the amount of quantum resources. This can serve as a way of simulating lattice gauge theories in high spatial dimensions, with non-Abelian gauge groups, and including dynamical fermions.

What happens when fermions hop on a lattice with crystalline defects? The answer depends on topological quantum numbers which specify the action of lattice rotations and translations in the low energy theory. One can understand the topological quantum numbers as a twist of continuum gauge fields in terms of crystalline gauge fields. We find that disclinations and dislocations -- defects of crystalline symmetries -- generally lead in the continuum to a certain ``emanant'' quantized magnetic flux. To demonstrate these facts, we study in detail tight-binding models whose low-energy descriptions are (2+1)D Dirac cones. Our map from lattice to continuum defects explains the crystalline topological response to disclinations and dislocations, and motivates the fermion crystalline equivalence principle used in the classification of crystalline topological phases. When the gap closes, the presence of emanant flux leads to pair creation from the vacuum with the particles and anti-particles swirling around the defect. We compute the associated currents and energy density using the tools of defect conformal field theory. There is a rich set of renormalization group fixed points, depending on how particles scatter from the defect. At half flux, there is a defect conformal manifold leading to a continuum of possible low-energy theories. We present extensive numerical evidence supporting the emanant magnetic flux at lattice defects and we test our map between lattice and continuum defects in detail. We also point out a no-go result, which implies that a single (2+1)D Dirac cone in symmetry class AII is incompatible with a commuting $C_M$ rotational symmetry with $(C_M)^M = +1$., Comment: 24+15 pages, 10+10 figures

Evaluating Retrieval-Augmented Generation (RAG) systems, especially in domain-specific contexts, requires benchmarks that address the distinctive requirements of the applicative scenario. Since real data can be hard to obtain, a common strategy is to use LLM-based methods to generate synthetic data. Existing solutions are general purpose: given a document, they generate a question to build a Q&A pair. However, although the generated questions can be individually good, they are typically not diverse enough to reasonably cover the different ways real end-users can interact with the RAG system. We introduce here DataMorgana, a tool for generating highly customizable and diverse synthetic Q&A benchmarks tailored to RAG applications. DataMorgana enables detailed configurations of user and question categories and provides control over their distribution within the benchmark. It uses a lightweight two-stage process, ensuring efficiency and fast iterations, while generating benchmarks that reflect the expected traffic. We conduct a thorough line of experiments, showing quantitatively and qualitatively that DataMorgana surpasses existing tools and approaches in producing lexically, syntactically, and semantically diverse question sets across domain-specific and general-knowledge corpora. DataMorgana will be made available to selected teams in the research community, as first beta testers, in the context of the upcoming SIGIR'2025 LiveRAG challenge to be announced in early February 2025.

Composite DNA is a recent novel method to increase the information capacity of DNA-based data storage above the theoretical limit of 2 bits/symbol. In this method, every composite symbol does not store a single DNA nucleotide but a mixture of the four nucleotides in a predetermined ratio. By using different mixtures and ratios, the alphabet can be extended to have much more than four symbols in the naive approach. While this method enables higher data content per synthesis cycle, potentially reducing the DNA synthesis cost, it also imposes significant challenges for accurate DNA sequencing since the base-level errors can easily change the mixture of bases and their ratio, resulting in changes to the composite symbols. With this motivation, we propose efficient constrained coding techniques to enforce the biological constraints, including the runlength-limited constraint and the GC-content constraint, into every DNA synthesized oligo, regardless of the mixture of bases in each composite letter and their corresponding ratio. Our goals include computing the capacity of the constrained channel, constructing efficient encoders/decoders, and providing the best options for the composite letters to obtain capacity-approaching codes. For certain codes' parameters, our methods incur only one redundant symbol.

Visual tokenization via auto-encoding empowers state-of-the-art image and video generative models by compressing pixels into a latent space. Although scaling Transformer-based generators has been central to recent advances, the tokenizer component itself is rarely scaled, leaving open questions about how auto-encoder design choices influence both its objective of reconstruction and downstream generative performance. Our work aims to conduct an exploration of scaling in auto-encoders to fill in this blank. To facilitate this exploration, we replace the typical convolutional backbone with an enhanced Vision Transformer architecture for Tokenization (ViTok). We train ViTok on large-scale image and video datasets far exceeding ImageNet-1K, removing data constraints on tokenizer scaling. We first study how scaling the auto-encoder bottleneck affects both reconstruction and generation -- and find that while it is highly correlated with reconstruction, its relationship with generation is more complex. We next explored the effect of separately scaling the auto-encoders' encoder and decoder on reconstruction and generation performance. Crucially, we find that scaling the encoder yields minimal gains for either reconstruction or generation, while scaling the decoder boosts reconstruction but the benefits for generation are mixed. Building on our exploration, we design ViTok as a lightweight auto-encoder that achieves competitive performance with state-of-the-art auto-encoders on ImageNet-1K and COCO reconstruction tasks (256p and 512p) while outperforming existing auto-encoders on 16-frame 128p video reconstruction for UCF-101, all with 2-5x fewer FLOPs. When integrated with Diffusion Transformers, ViTok demonstrates competitive performance on image generation for ImageNet-1K and sets new state-of-the-art benchmarks for class-conditional video generation on UCF-101., Comment: 28 pages, 25 figures, 7 Tables

Quantum sensors offer exceptional sensitivity to nanoscale magnetic field fluctuations, where non-stationary effects such as spin diffusion and non-Markovian dynamics arising from coupling to few environmental degrees of freedom play critical roles. Here, we demonstrate how quantum sensors can characterize the statistical properties of noise sources, distinguishing between stationary and non-stationary behaviors, as well as Markovian and non-Markovian dynamics. Using nitrogen-vacancy (NV) centers in diamond as a platform, we develop a physical noise model that analytically predicts Ramsey decay curves under different noise regimes. These predictions are experimentally validated by measuring Ramsey decay for NV centers subject to injected noise of each type. Our results showcase the capability of quantum sensors to unravel complex noise behaviors induced by nanoscale environments, shedding light on their physical origins and guiding the development of strategies to mitigate decoherence. This work deepens our understanding of noise dynamics at the nanoscale and lays the foundation for enhancing the performance and robustness of quantum technologies.

We consider the task of Image-to-Video (I2V) generation, which involves transforming static images into realistic video sequences based on a textual description. While recent advancements produce photorealistic outputs, they frequently struggle to create videos with accurate and consistent object motion, especially in multi-object scenarios. To address these limitations, we propose a two-stage compositional framework that decomposes I2V generation into: (i) An explicit intermediate representation generation stage, followed by (ii) A video generation stage that is conditioned on this representation. Our key innovation is the introduction of a mask-based motion trajectory as an intermediate representation, that captures both semantic object information and motion, enabling an expressive but compact representation of motion and semantics. To incorporate the learned representation in the second stage, we utilize object-level attention objectives. Specifically, we consider a spatial, per-object, masked-cross attention objective, integrating object-specific prompts into corresponding latent space regions and a masked spatio-temporal self-attention objective, ensuring frame-to-frame consistency for each object. We evaluate our method on challenging benchmarks with multi-object and high-motion scenarios and empirically demonstrate that the proposed method achieves state-of-the-art results in temporal coherence, motion realism, and text-prompt faithfulness. Additionally, we introduce \benchmark, a new challenging benchmark for single-object and multi-object I2V generation, and demonstrate our method's superiority on this benchmark. Project page is available at https://guyyariv.github.io/TTM/.

Lattice gauge theory is an important framework for studying gauge theories that arise in the Standard Model and condensed matter physics. Yet many systems (or regimes of those systems) are difficult to study using conventional techniques, such as action-based Monte Carlo sampling. In this paper, we demonstrate the use of gauged Gaussian projected entangled pair states as an ansatz for a lattice gauge theory involving dynamical physical matter. We study a $\mathbb{Z}_2$ gauge theory on a two dimensional lattice with a single flavor of fermionic matter on each lattice site. Our results show agreement with results computed by exactly diagonalizing the Hamiltonian, and demonstrate that the approach is computationally feasible for larger system sizes where exact results are unavailable. This is a further step on the road to studying higher dimensions and other gauge groups with manageable computational costs while avoiding the sign problem., Comment: 8 figures

Despite the rapid integration of video perception capabilities into Large Multimodal Models (LMMs), the underlying mechanisms driving their video understanding remain poorly understood. Consequently, many design decisions in this domain are made without proper justification or analysis. The high computational cost of training and evaluating such models, coupled with limited open research, hinders the development of video-LMMs. To address this, we present a comprehensive study that helps uncover what effectively drives video understanding in LMMs. We begin by critically examining the primary contributors to the high computational requirements associated with video-LMM research and discover Scaling Consistency, wherein design and training decisions made on smaller models and datasets (up to a critical size) effectively transfer to larger models. Leveraging these insights, we explored many video-specific aspects of video-LMMs, including video sampling, architectures, data composition, training schedules, and more. For example, we demonstrated that fps sampling during training is vastly preferable to uniform frame sampling and which vision encoders are the best for video representation. Guided by these findings, we introduce Apollo, a state-of-the-art family of LMMs that achieve superior performance across different model sizes. Our models can perceive hour-long videos efficiently, with Apollo-3B outperforming most existing $7$B models with an impressive 55.1 on LongVideoBench. Apollo-7B is state-of-the-art compared to 7B LMMs with a 70.9 on MLVU, and 63.3 on Video-MME., Comment: https://apollo-lmms.github.io

High temperatures are typically thought to increase disorder. Here we examine this idea in Quantum Field Theory in 2+1 dimensions. For this sake we explore a novel class of tractable models, consisting of nearly-mean-field scalars interacting with critical scalars. We identify UV-complete, local, unitary models in this class and show that symmetry breaking $\mathbb{Z}_2 \to \emptyset$ occurs at any temperature in some regions of the phase diagram. This phenomenon, previously observed in models with fractional dimensions, or in the strict planar limits, or with non-local interactions, is now exhibited in a local, unitary 2+1 dimensional model with a finite number of fields., Comment: 8 pages, 1 figure

Storing data in DNA is being explored as an efficient solution for archiving and in-object storage. Synthesis time and cost remain challenging, significantly limiting some applications at this stage. In this paper we investigate efficient synthesis, as it relates to cyclic synchronized synthesis technologies, such as photolithography. We define performance metrics related to the number of cycles needed for the synthesis of any fixed number of bits. We first expand on some results from the literature related to the channel capacity, addressing densities beyond those covered by prior work. This leads us to develop effective encoding achieving rate and capacity that are higher than previously reported. Finally, we analyze cost based on a parametric definition and determine some bounds and asymptotics. We investigate alphabet sizes that can be larger than 4, both for theoretical completeness and since practical approaches to such schemes were recently suggested and tested in the literature.

The application of deep learning methods, particularly foundation models, in biological research has surged in recent years. These models can be text-based or trained on underlying biological data, especially omics data of various types. However, comparing the performance of these models consistently has proven to be a challenge due to differences in training data and downstream tasks. To tackle this problem, we developed an architecture-agnostic benchmarking approach that, instead of evaluating the models directly, leverages entity representation vectors from each model and trains simple predictive models for each benchmarking task. This ensures that all types of models are evaluated using the same input and output types. Here we focus on gene properties collected from professionally curated bioinformatics databases. These gene properties are categorized into five major groups: genomic properties, regulatory functions, localization, biological processes, and protein properties. Overall, we define hundreds of tasks based on these databases, which include binary, multi-label, and multi-class classification tasks. We apply these benchmark tasks to evaluate expression-based models, large language models, protein language models, DNA-based models, and traditional baselines. Our findings suggest that text-based models and protein language models generally outperform expression-based models in genomic properties and regulatory functions tasks, whereas expression-based models demonstrate superior performance in localization tasks. These results should aid in the development of more informed artificial intelligence strategies for biological understanding and therapeutic discovery. To ensure the reproducibility and transparency of our findings, we have made the source code and benchmark data publicly accessible for further investigation and expansion at github.com/BiomedSciAI/gene-benchmark.

Conventionally, a first-in-human phase I trial in healthy volunteers aims to confirm the safety of a drug in humans. In such situations, volunteers should not suffer from any safety issues and simple algorithm-based dose-escalation schemes are often used. However, to avoid too many clinical trials in the future, it might be appealing to design these trials to accumulate information on the link between dose and efficacy/activity under strict safety constraints. Furthermore, an increasing number of molecules for which the increasing dose-activity curve reaches a plateau are emerging.In a phase I dose-finding trial context, our objective is to determine, under safety constraints, among a set of doses, the lowest dose whose probability of activity is closest to a given target. For this purpose, we propose a two-stage dose-finding design. The first stage is a typical algorithm dose escalation phase that can both check the safety of the doses and accumulate activity information. The second stage is a model-based dose-finding phase that involves selecting the best dose-activity model according to the plateau location.Our simulation study shows that our proposed method performs better than the common Bayesian logistic regression model in selecting the optimal dose.

Gauged Gaussian fermionic projected entangled pair states (GGFPEPS) form a novel type of Ansatz state for the groundstate of lattice gauge theories. The advantage of these states is that they allow efficient calculation of observables by combining Monte-Carlo integration over gauge fields configurations with Gaussian tensor network machinery for the fermionic part. Remarkably, for GGFPEPS the probability distribution for the gauge field configurations is positive definite and real so that there is no sign problem. In this work we will demonstrate that gauged (non-Gaussian) fermionic projected pair states (GFPEPS) exactly capture the groundstate of generic lattice gauge theories. Additionally, we will present a framework for the efficient computation of observables in the case where the non-Gaussianity of the PEPS follows from the superposition of (few) Gaussian PEPS. Finally, we present a new graphical notation for Gaussian tensor and their contractions into Gaussian tensor network states.

Different choices of quantum error-correcting codes can reduce the demands on the physical hardware needed to build a quantum computer. To achieve the full potential of a code, we must develop practical decoding algorithms that can correct errors that have occurred with high likelihood. Matching decoders are very good at correcting local errors while also demonstrating fast run times that can keep pace with physical quantum devices. We implement variations of a matching decoder for a three-dimensional, non-CSS, low-density parity check code known as the Chamon code, which has a non-trivial structure that does not lend itself readily to this type of decoding. The non-trivial structure of the syndrome of this code means that we can supplement the decoder with additional steps to improve the threshold error rate, below which the logical failure rate decreases with increasing code distance. We find that a generalized matching decoder that is augmented by a belief-propagation step prior to matching gives a threshold of 10.5% for depolarising noise., Comment: 9 pages, 7 figures

The phase diagram of QCD may contain a moat regime in a large region of temperature $T$ and chemical potential $\mu\neq0$. A moat regime is characterized by quasiparticle moatons (pions) whose energy is minimal at nonzero spatial momentum. At $\mu\neq 0$, higher mass dimension operators play a critical role in a moat regime. At dimension six, there are nine possible gauge invariant couplings between scalars and photons. For back-to-back dilepton production, only one operator contributes, which significantly enhances production near a moat threshold. This enhancement is an experimental signature of moatons., Comment: 9 pages, 3 figures; 4 pages references & supplemental material

Projected entangled pair states (PEPS) are very useful in the description of strongly correlated systems, partly because they allow encoding symmetries, either global or local (gauge), naturally. In recent years, PEPS with local symmetries have increasingly been used in the study of non-perturbative regimes of lattice gauge theories, most prominently as a way to construct variational ansatz states depending only on a small number of parameters and yet capturing the relevant physical properties. For the case of one-dimensional PEPS (Matrix Product States - MPS) a bidirectional connection was established between the internal structure of the tensor network, i.e. the mathematical properties of the constituent tensors, and the symmetry. In higher dimensions this has only been done for global symmetries, where in the local (gauge) case it is known only how to construct gauge-invariant states, but not what the symmetry implies on the internal structure of the PEPS. In the present work we complete this missing piece and study the internal structure of projected entangled pair states with a gauge symmetry. The PEPS we consider consist of matter and gauge field tensors placed on the vertices and edges, respectively, of arbitrary graphs., Comment: 24 pages, 4 figures; included example section

We present Movie Gen, a cast of foundation models that generates high-quality, 1080p HD videos with different aspect ratios and synchronized audio. We also show additional capabilities such as precise instruction-based video editing and generation of personalized videos based on a user's image. Our models set a new state-of-the-art on multiple tasks: text-to-video synthesis, video personalization, video editing, video-to-audio generation, and text-to-audio generation. Our largest video generation model is a 30B parameter transformer trained with a maximum context length of 73K video tokens, corresponding to a generated video of 16 seconds at 16 frames-per-second. We show multiple technical innovations and simplifications on the architecture, latent spaces, training objectives and recipes, data curation, evaluation protocols, parallelization techniques, and inference optimizations that allow us to reap the benefits of scaling pre-training data, model size, and training compute for training large scale media generation models. We hope this paper helps the research community to accelerate progress and innovation in media generation models. All videos from this paper are available at https://go.fb.me/MovieGenResearchVideos.

Quantum simulation is one of the methods that have been proposed and used in practice to bypass computational challenges in the investigation of lattice gauge theories. While most of the proposals rely on truncating the infinite dimensional Hilbert spaces that these models feature, we propose a truncation-free method based on the exact analogy between the local Hilbert space of lattice QED and that of a Josephson junction. We provide several proposals, mostly semi-analog, arranged according to experimental difficulty. Our method can simulate a quasi-2D system of up to $2\times N$ plaquettes, and we present an approximate method that can simulate the fully-2D theory, but is more demanding experimentally and not immediately feasible. This sets the ground for analog quantum simulation of lattice gauge theories with superconducting circuits, in a completely Hilbert space truncation-free procedure, for continuous gauge groups., Comment: 12 pages, 6 figures. v2: added numerical estimation of errors

DNA-based data storage has been attracting significant attention due to its extremely high data storage density, low power consumption, and long duration compared to conventional data storage media. Despite the recent advancements in DNA data storage technology, significant challenges remain. In particular, various types of errors can occur during the processes of DNA synthesis, storage, and sequencing, including substitution errors, insertion errors, and deletion errors. Furthermore, the entire oligo may be lost. In this work, we report a DNA-based data storage architecture that incorporates efficient channel coding schemes, including different types of error-correcting codes (ECCs) and constrained codes, for both the inner coding and outer coding for the DNA data storage channel. We also carried out large scale experiments to validate our proposed DNA-based data storage architecture. Specifically, 1.61 and 1.69 MB data were encoded into 30,000 oligos each, with information densities of 1.731 and 1.815, respectively. It has been found that the stored information can be fully recovered without any error at average coverages of 4.5 and 6.0, respectively. This experiment achieved the highest net information density and lowest coverage among existing DNA-based data storage experiments (with standard DNA), with data recovery rates and coverage approaching theoretical optima.

Quantum Electrodynamics in 2+1 dimensions (QED$_3$) with two Dirac fermions displays time reversal symmetry, nontrivial SPT phases and anomalies. The fate of this theory in its strongly coupled regime has been debated extensively. Surprisingly, we find that gluing together the phase diagrams of two standard Wilson-Fisher $O(4)$ theories suffices to reproduce all the SPT phases, anomalies, and semi-classical limits. A central mechanism behind it is ``SPT absorption''. The patching of the $O(4)$ transitions makes very concrete predictions for the behavior of the theory in its strongly coupled limits; for instance, the $\theta=\pi$ sigma model with $S^3$ topology appears due to monopole condensation., Comment: 6 pages, 3 figures, v2 submitted for publication

Training large language models (LLMs) for external tool usage is a rapidly expanding field, with recent research focusing on generating synthetic data to address the shortage of available data. However, the absence of systematic data quality checks poses complications for properly training and testing models. To that end, we propose two approaches for assessing the reliability of data for training LLMs to use external tools. The first approach uses intuitive, human-defined correctness criteria. The second approach uses a model-driven assessment with in-context evaluation. We conduct a thorough evaluation of data quality on two popular benchmarks, followed by an extrinsic evaluation that showcases the impact of data quality on model performance. Our results demonstrate that models trained on high-quality data outperform those trained on unvalidated data, even when trained with a smaller quantity of data. These findings empirically support the significance of assessing and ensuring the reliability of training data for tool-using LLMs.

Quantum query complexity mainly studies the number of queries needed to learn some property of a black box with high probability. A closely related question is how well an algorithm can succeed with this learning task using only a fixed number of queries. In this work, we propose measuring an algorithm's performance using the mutual information between the output and the actual value. A key observation is that if an algorithm is only allowed to make a single query and the goal is to optimize this mutual information, then we obtain a task which is similar to a basic task of quantum communication, where one attempts to maximize the mutual information of the sender and receiver. We make this analogy precise by formally considering the oracle as a separate subsystem, whose state records the unknown oracle identity. The oracle query prepares a state, which is then measured; and the target property of the oracle plays the role of a message that should be deduced from the measurement outcome. Thus we obtain a link between the optimal single-query algorithm and minimization of the extent of quantum correlations between the oracle and the computer subsystems. We also find a lower bound on this mutual information, which is related to quantum coherence. These results extend to multiple-query non-adaptive algorithms. As a result, we gain insight into the task of finding the optimal non-adaptive algorithm that uses at most a fixed number of queries, for any oracle problem., Comment: 19 pages, 1 figure, 5 tables

This work investigates the use of a fiber piano in controlling spatial modes in few mode fibers. It has been found that together with sub-optimal coupling into SMF-28 fibre and half and quarter waveplates, the fiber piano is capable of producing and reproducing desired spatial modes up to $LP_{11}$ when using 808 nm light and up to $LP_{21}$ when using 632.8 nm light. The control of spatial mode profile extends down to the single photon level. This is demonstrated with the help of correlated photon pairs generated via spontaneous parametric down conversion., Comment: 7 pages, 7 figures

The performance of Large Vision Language Models (LVLMs) is dependent on the size and quality of their training datasets. Existing video instruction tuning datasets lack diversity as they are derived by prompting large language models with video captions to generate question-answer pairs, and are therefore mostly descriptive. Meanwhile, many labeled video datasets with diverse labels and supervision exist - however, we find that their integration into LVLMs is non-trivial. Herein, we present Video Self-Training with augmented Reasoning (Video-STaR), the first video self-training approach. Video-STaR allows the utilization of any labeled video dataset for video instruction tuning. In Video-STaR, an LVLM cycles between instruction generation and finetuning, which we show (I) improves general video understanding and (II) adapts LVLMs to novel downstream tasks with existing supervision. During generation, an LVLM is prompted to propose an answer. The answers are then filtered only to those that contain the original video labels, and the LVLM is then re-trained on the generated dataset. By only training on generated answers that contain the correct video labels, Video-STaR utilizes these existing video labels as weak supervision for video instruction tuning. Our results demonstrate that Video-STaR-enhanced LVLMs exhibit improved performance in (I) general video QA, where TempCompass performance improved by 10%, and (II) on downstream tasks, where Video-STaR improved Kinetics700-QA accuracy by 20% and action quality assessment on FineDiving by 15%., Comment: Project page: https://orrzohar.github.io/projects/video-star/

Stable infiniteness, strong finite witnessability, and smoothness are model-theoretic properties relevant to theory combination in satisfiability modulo theories. Theories that are strongly finitely witnessable and smooth are called strongly polite and can be effectively combined with other theories. Toledo, Zohar, and Barrett conjectured that stably infinite and strongly finitely witnessable theories are smooth and therefore strongly polite. They called counterexamples to this conjecture unicorn theories, as their existence seemed unlikely. We prove that, indeed, unicorns do not exist. We also prove versions of the L\"owenheim-Skolem theorem and the {\L}o\'s-Vaught test for many-sorted logic., Comment: To appear in FM24

Large Language Models (LLMs) and strong vision models have enabled rapid research and development in the field of Vision-Language-Action models that enable robotic control. The main objective of these methods is to develop a generalist policy that can control robots with various embodiments. However, in industrial robotic applications such as automated assembly and disassembly, some tasks, such as insertion, demand greater accuracy and involve intricate factors like contact engagement, friction handling, and refined motor skills. Implementing these skills using a generalist policy is challenging because these policies might integrate further sensory data, including force or torque measurements, for enhanced precision. In our method, we present a global control policy based on LLMs that can transfer the control policy to a finite set of skills that are specifically trained to perform high-precision tasks through dynamic context switching. The integration of LLMs into this framework underscores their significance in not only interpreting and processing language inputs but also in enriching the control mechanisms for diverse and intricate robotic operations.

Generative methods have recently seen significant improvements by generating in a lower-dimensional latent representation of the data. However, many of the generative methods applied in the latent space remain complex and difficult to train. Further, it is not entirely clear why transitioning to a lower-dimensional latent space can improve generative quality. In this work, we introduce a new and simple generative method grounded in topology theory -- Generative Topological Networks (GTNs) -- which also provides insights into why lower-dimensional latent-space representations might be better-suited for data generation. GTNs are simple to train -- they employ a standard supervised learning approach and do not suffer from common generative pitfalls such as mode collapse, posterior collapse or the need to pose constraints on the neural network architecture. We demonstrate the use of GTNs on several datasets, including MNIST, CelebA, CIFAR-10 and the Hands and Palm Images dataset by training GTNs on a lower-dimensional latent representation of the data. We show that GTNs can improve upon VAEs and that they are quick to converge, generating realistic samples in early epochs. Further, we use the topological considerations behind the development of GTNs to offer insights into why generative models may benefit from operating on a lower-dimensional latent space, highlighting the important link between the intrinsic dimension of the data and the dimension in which the data is generated. Particularly, we demonstrate that generating in high dimensional ambient spaces may be a contributing factor to out-of-distribution samples generated by diffusion models. We also highlight other topological properties that are important to consider when using and designing generative models. Our code is available at: https://github.com/alonalj/GTN

In CFTs, the partition function of a line defect with a cusp depends logarithmically on the size of the line with an angle-dependent coefficient: the cusp anomalous dimension. In the first part of this work, we study the general properties of the cusp anomalous dimension. We relate the small cusp angle limit to the effective field theory of defect fusion, making predictions for the first couple of terms in the expansion. Using a concavity property of the cusp anomalous dimension we argue that the Casimir energy between a line defect and its orientation reversal is always negative ("opposites attract"). We use these results to determine the fusion algebra of Wilson lines in $\mathcal{N}=4$ SYM as well as pinning field defects in the Wilson-Fisher fixed points. In the second part of the paper we obtain nonperturbative numerical results for the cusp anomalous dimension of pinning field defects in the Ising model in $d=3$, using the recently developed fuzzy-sphere regularization. We also compute the pinning field cusp anomalous dimension in the $O(N)$ model at one-loop in the $\varepsilon$-expansion. Our results are in agreement with the general theory developed in the first part of the work, and we make several predictions for impurities in magnets., Comment: 35 pages + appendices, 15 figures; v2 typos fixed; v3 journal version

The combination of Monte Carlo tree search and neural networks has revolutionized online planning. As neural network approximations are often imperfect, we ask whether uncertainty estimates about the network outputs could be used to improve planning. We develop a Bayesian planning approach that facilitates such uncertainty quantification, inspired by classical ideas from the meta-reasoning literature. We propose a Thompson sampling based algorithm for searching the tree of possible actions, for which we prove the first (to our knowledge) finite time Bayesian regret bound, and propose an efficient implementation for a restricted family of posterior distributions. In addition we propose a variant of the Bayes-UCB method applied to trees. Empirically, we demonstrate that on the ProcGen Maze and Leaper environments, when the uncertainty estimates are accurate but the neural network output is inaccurate, our Bayesian approach searches the tree much more effectively. In addition, we investigate whether popular uncertainty estimation methods are accurate enough to yield significant gains in planning. Our code is available at: https://github.com/nirgreshler/bayesian-online-planning.

Kernel ridge regression (KRR) and Gaussian processes (GPs) are fundamental tools in statistics and machine learning, with recent applications to highly over-parameterized deep neural networks. The ability of these tools to learn a target function is directly related to the eigenvalues of their kernel sampled on the input data distribution. Targets that have support on higher eigenvalues are more learnable. However, solving such eigenvalue problems on real-world data remains a challenge. Here, we consider cross-dataset learnability and show that one may use eigenvalues and eigenfunctions associated with highly idealized data measures to reveal spectral bias on complex datasets and bound learnability on real-world data. This allows us to leverage various symmetries that realistic kernels manifest to unravel their spectral bias.