Search

Given the rapid progress of generative AI, there is a pressing need to systematically compare and choose between the numerous models and configurations available. The scale and versatility of such evaluations make the use of LLM-based judges a compelling solution for this challenge. Crucially, this approach requires first to validate the quality of the LLM judge itself. Previous work has focused on instance-based assessment of LLM judges, where a judge is evaluated over a set of responses, or response pairs, while being agnostic to their source systems. We argue that this setting overlooks critical factors affecting system-level ranking, such as a judge's positive or negative bias towards certain systems. To address this gap, we conduct the first large-scale study of LLM judges as system rankers. System scores are generated by aggregating judgment scores over multiple system outputs, and the judge's quality is assessed by comparing the resulting system ranking to a human-based ranking. Beyond overall judge assessment, our analysis provides a fine-grained characterization of judge behavior, including their decisiveness and bias.

Although pandemics are often studied as if populations are well-mixed, disease transmission networks exhibit a multi-scale structure stretching from the individual all the way up to the entire globe. The COVID-19 pandemic has led to an intense debate about whether interventions should prioritize public health or the economy, leading to a surge of studies analyzing the health and economic costs of various response strategies. Here we show that describing disease transmission in a self-similar (fractal) manner across multiple geographic scales allows for the design of multi-scale containment measures that substantially reduce both these costs. We characterize response strategies using multi-scale reproduction numbers -- a generalization of the basic reproduction number $R_0$ -- that describe pandemic spread at multiple levels of scale and provide robust upper bounds on disease transmission. Stable elimination is guaranteed if there exists a scale such that the reproduction number among regions of that scale is less than $1$, even if the basic reproduction number $R_0$ is greater than $1$. We support our theoretical results using simulations of a heterogeneous SIS model for disease spread in the United States constructed using county-level commuting, air travel, and population data.

We present ZeroBAS, a neural method to synthesize binaural audio from monaural audio recordings and positional information without training on any binaural data. To our knowledge, this is the first published zero-shot neural approach to mono-to-binaural audio synthesis. Specifically, we show that a parameter-free geometric time warping and amplitude scaling based on source location suffices to get an initial binaural synthesis that can be refined by iteratively applying a pretrained denoising vocoder. Furthermore, we find this leads to generalization across room conditions, which we measure by introducing a new dataset, TUT Mono-to-Binaural, to evaluate state-of-the-art monaural-to-binaural synthesis methods on unseen conditions. Our zero-shot method is perceptually on-par with the performance of supervised methods on the standard mono-to-binaural dataset, and even surpasses them on our out-of-distribution TUT Mono-to-Binaural dataset. Our results highlight the potential of pretrained generative audio models and zero-shot learning to unlock robust binaural audio synthesis.

We prove that every small profinite group can be decomposed into a direct product of indecomposable profinite groups, and that such a decomposition is unique up to order and isomorphisms of the components. We also investigate the cancellation property of some free pro-$\mathcal{C}$ groups, and give a new criterion for a profinite group to be small.

We investigate the problem of selecting features for datasets that can be naturally partitioned into subgroups (e.g., according to socio-demographic groups and age), each with its own dominant set of features. Within this subgroup-oriented framework, we address the challenge of systematic missing data, a scenario in which some feature values are missing for all tuples of a subgroup, due to flawed data integration, regulatory constraints, or privacy concerns. Feature selection is governed by finding mutual Information, a popular quantification of correlation, between features and a target variable. Our goal is to identify top-K feature subsets of some fixed size with the highest joint mutual information with a target variable. In the presence of systematic missing data, the closed form of mutual information could not simply be applied. We argue that in such a setting, leveraging relationships between available feature mutual information within a subgroup or across subgroups can assist inferring missing mutual information values. We propose a generalizable model based on heterogeneous graph neural network to identify interdependencies between feature-subgroup-target variable connections by modeling it as a multiplex graph, and employing information propagation between its nodes. We address two distinct scalability challenges related to training and propose principled solutions to tackle them. Through an extensive empirical evaluation, we demonstrate the efficacy of the proposed solutions both qualitatively and running time wise.

People who are blind perceive the world differently than those who are sighted, which can result in distinct motion characteristics. For instance, when crossing at an intersection, blind individuals may have different patterns of movement, such as veering more from a straight path or using touch-based exploration around curbs and obstacles. These behaviors may appear less predictable to motion models embedded in technologies such as autonomous vehicles. Yet, the ability of 3D motion models to capture such behavior has not been previously studied, as existing datasets for 3D human motion currently lack diversity and are biased toward people who are sighted. In this work, we introduce BlindWays, the first multimodal motion benchmark for pedestrians who are blind. We collect 3D motion data using wearable sensors with 11 blind participants navigating eight different routes in a real-world urban setting. Additionally, we provide rich textual descriptions that capture the distinctive movement characteristics of blind pedestrians and their interactions with both the navigation aid (e.g., a white cane or a guide dog) and the environment. We benchmark state-of-the-art 3D human prediction models, finding poor performance with off-the-shelf and pre-training-based methods for our novel task. To contribute toward safer and more reliable systems that can seamlessly reason over diverse human movements in their environments, our text-and-motion benchmark is available at https://blindways.github.io., Comment: Accepted at NeurIPS 2024

Navigation is a fundamental skill of agents with visual-motor capabilities. We introduce a Navigation World Model (NWM), a controllable video generation model that predicts future visual observations based on past observations and navigation actions. To capture complex environment dynamics, NWM employs a Conditional Diffusion Transformer (CDiT), trained on a diverse collection of egocentric videos of both human and robotic agents, and scaled up to 1 billion parameters. In familiar environments, NWM can plan navigation trajectories by simulating them and evaluating whether they achieve the desired goal. Unlike supervised navigation policies with fixed behavior, NWM can dynamically incorporate constraints during planning. Experiments demonstrate its effectiveness in planning trajectories from scratch or by ranking trajectories sampled from an external policy. Furthermore, NWM leverages its learned visual priors to imagine trajectories in unfamiliar environments from a single input image, making it a flexible and powerful tool for next-generation navigation systems., Comment: project page: https://www.amirbar.net/nwm/

In the study of animal behavior, researchers often record long continuous videos, accumulating into large-scale datasets. However, the behaviors of interest are often rare compared to routine behaviors. This incurs a heavy cost on manual annotation, forcing users to sift through many samples before finding their needles. We propose a pipeline to efficiently sample rare behaviors from large datasets, enabling the creation of training datasets for rare behavior classifiers. Our method only needs an unlabeled animal pose or acceleration dataset as input and makes no assumptions regarding the type, number, or characteristics of the rare behaviors. Our pipeline is based on a recent graph-based anomaly detection model for human behavior, which we apply to this new data domain. It leverages anomaly scores to automatically label normal samples while directing human annotation efforts toward anomalies. In research data, anomalies may come from many different sources (e.g., signal noise versus true rare instances). Hence, the entire labeling budget is focused on the abnormal classes, letting the user review and label samples according to their needs. We tested our approach on three datasets of freely-moving animals, acquired in the laboratory and the field. We found that graph-based models are particularly useful when studying motion-based behaviors in animals, yielding good results while using a small labeling budget. Our method consistently outperformed traditional random sampling, offering an average improvement of 70% in performance and creating datasets even when the behavior of interest was only 0.02% of the data. Even when the performance gain was minor (e.g., when the behavior is not rare), our method still reduced the annotation effort by half.

Advancements in text-to-image diffusion models have led to significant progress in fast 3D content creation. One common approach is to generate a set of multi-view images of an object, and then reconstruct it into a 3D model. However, this approach bypasses the use of a native 3D representation of the object and is hence prone to geometric artifacts and limited in controllability and manipulation capabilities. An alternative approach involves native 3D generative models that directly produce 3D representations. These models, however, are typically limited in their resolution, resulting in lower quality 3D objects. In this work, we bridge the quality gap between methods that directly generate 3D representations and ones that reconstruct 3D objects from multi-view images. We introduce a multi-view to multi-view diffusion model called Sharp-It, which takes a 3D consistent set of multi-view images rendered from a low-quality object and enriches its geometric details and texture. The diffusion model operates on the multi-view set in parallel, in the sense that it shares features across the generated views. A high-quality 3D model can then be reconstructed from the enriched multi-view set. By leveraging the advantages of both 2D and 3D approaches, our method offers an efficient and controllable method for high-quality 3D content creation. We demonstrate that Sharp-It enables various 3D applications, such as fast synthesis, editing, and controlled generation, while attaining high-quality assets., Comment: Project page at https://yiftachede.github.io/Sharp-It/

Active Learning (AL) is a user-interactive approach aimed at reducing annotation costs by selecting the most crucial examples to label. Although AL has been extensively studied for image classification tasks, the specific scenario of interactive image retrieval has received relatively little attention. This scenario presents unique characteristics, including an open-set and class-imbalanced binary classification, starting with very few labeled samples. We introduce a novel batch-mode Active Learning framework named GAL (Greedy Active Learning) that better copes with this application. It incorporates a new acquisition function for sample selection that measures the impact of each unlabeled sample on the classifier. We further embed this strategy in a greedy selection approach, better exploiting the samples within each batch. We evaluate our framework with both linear (SVM) and non-linear MLP/Gaussian Process classifiers. For the Gaussian Process case, we show a theoretical guarantee on the greedy approximation. Finally, we assess our performance for the interactive content-based image retrieval task on several benchmarks and demonstrate its superiority over existing approaches and common baselines. Code is available at https://github.com/barleah/GreedyAL., Comment: Accepted to Transactions on Machine Learning Research (TMLR)

An alternative to the dependence on traditional student loans may offer a viable relief from the tremendous burden that those loans usually incur. This article establishes that it is desirable for governmental intervention to grant students 'more choice' in their funding decisions by allowing them to have portfolios, mixtures of different types of loans. To emphasize this point, a model is presented of a situation where students invest in higher education while facing uncertainty about their individual earning potential. The model reveals that when students are allowed to have portfolios of loans, some of them indeed take the opportunity and diversify their loans, benefiting themselves, but also improving the loan terms of other students. Therefore, when governments organize student loans, they should consider providing students with more choice in their funding decisions.

Diamond single crystals could be an attractive nonlinear THz source due to the materials high damage threshold, high transparency and THz-NIR phase matching ability. However, the buildup of second order nonlinear fields is restricted due to the centrosymmetric structure of the crystal. Here, we demonstrate nonlinear broadband THz emission due to optical rectification from diamond single crystals induced by embedded symmetry breaking nitrogen vacancy (NV) centers. The nonlinear temporal and spectral emission properties of the crystal are examined using THz time domain spectroscopy after pumping with NIR femtosecond pulses. The diamond-NV cell structure was also examined using density functional theory (DFT). The results show the potential of NV centers in diamond as a new nonlinear platform for broadband efficient THz generation., Comment: 8 pages, 3 figures

How diverse are the outputs of large language models when diversity is desired? We examine the diversity of responses of various models to questions with multiple possible answers, comparing them with human responses. Our findings suggest that models' outputs are highly concentrated, reflecting a narrow, mainstream 'worldview', in comparison to humans, whose responses exhibit a much longer-tail. We examine three ways to increase models' output diversity: 1) increasing generation randomness via temperature sampling; 2) prompting models to answer from diverse perspectives; 3) aggregating outputs from several models. A combination of these measures significantly increases models' output diversity, reaching that of humans. We discuss implications of these findings for AI policy that wishes to preserve cultural diversity, an essential building block of a democratic social fabric.

We study the collective dissipative dynamics of dipoles modeled as harmonic oscillators coupled to 1-D electromagnetic reservoirs. The bosonic nature of the dipole oscillators as well as the reservoir modes allows an exact numerical simulation of the dynamics for arbitrary coupling strengths. At weak coupling, apart from essentially recovering the dynamics expected from a Markovian Lindblad master equation, we also obtain non-Markovian effects for spatially separated two-level emitters. In the so called ultrastrong coupling regime, we find the dynamics and steady state depends on the choice of the reservoir which is chosen as either an ideal cavity with equispaced, unbounded dispersion or a cavity array with a bounded dispersion. Moreover, at even higher coupling strengths, we find a decoupling between the light and matter degrees of freedom attributable to the increased importance of the diamagnetic term in the Hamiltonian. In this regime, we find that the dependence of the dynamics on the separation between the dipoles is not important and the dynamics is dominated by the occupation of the polariton mode of lowest energy.

The proton capture cross-section of the least abundant proton-rich stable isotope of cadmium, $^{108}$Cd (abundance 0.89\%), has been measured near the Gamow window corresponding to a temperature range of 3-4 GK. The measurement of the $^{108}$Cd(p,$\gamma$)$^{109}$In reaction was carried out using the activation technique. The cross-section at the lowest energy point of 3T$_9$, E$_p$$^{lab}$= 2.28 MeV, has been reported for the first time. The astrophysical S-factor was measured in the energy range relevant to the astrophysical p-process, between E$_p$$^{cm}$= 2.29 and 6.79 MeV. The experimental results have been compared with theoretical predictions of Hauser-Feshbach statistical model calculations using TALYS-1.96. A calculated proton-optical potential was implemented to achieve better fitting, with different combinations of available nuclear level densities (NLDs) and $\gamma$-ray strength functions in TALYS-1.96. The calculations provided satisfactory agreement with the experimental results. The reaction rate was calculated using the calculated potential in TALYS-1.96 and compared with the values provided in the REACLIB database.

We investigate the internal representations of vision-and-language models (VLMs) and how they encode task representations. We consider tasks specified through examples or instructions, using either text or image inputs. Surprisingly, we find that conceptually similar tasks are mapped to similar task vector representations, regardless of how they are specified. Our findings suggest that to output answers, tokens in VLMs undergo three distinct phases: input, task, and answer, a process which is consistent across different modalities and specifications. The task vectors we identify in VLMs are general enough to be derived in one modality (e.g., text) and transferred to another (e.g., image). Additionally, we find that ensembling exemplar and instruction based task vectors produce better task representations. Taken together, these insights shed light on the underlying mechanisms of VLMs, particularly their ability to represent tasks in a shared manner across different modalities and task specifications. Project page: https://task-vectors-are-cross-modal.github.io.

Metabolite biosynthesis is regulated via metabolic pathways, which can be activated and deactivated within organisms. Understanding and identifying an organism's metabolic pathway network is a crucial aspect for various research fields, including crop and life stock breeding, pharmacology, and medicine. The problem of identifying whether a pathway is part of a studied metabolic system is commonly framed as a hyperlink prediction problem. The most important challenge in prediction of metabolic pathways is the sparsity of the labeled data. This challenge can partially be mitigated using metabolite correlation networks which are affected by all active pathways including those that were not confirmed yet in laboratory experiments. Unfortunately, extracting properties that can confirm or refute existence of a metabolic pathway in a particular organism is not a trivial task. In this research, we introduce the Network Auralization Hyperlink Prediction (NetAurHPD) which is a framework that relies on (1) graph auralization to extract and aggregate representations of nodes in metabolite correlation networks and (2) data augmentation method that generates metabolite correlation networks given a subset of chemical reactions defined as hyperlinks. Experiments with metabolites correlation-based networks of tomato pericarp demonstrate promising results for NetAurHPD, compared to alternative methods. Furthermore, the application of data augmentation improved NetAurHPD's learning capabilities and overall performance. Additionally, NetAurHPD outperformed state-of-the-art method in experiments under challenging conditions, and has the potential to be a valuable tool for exploring organisms with limited existing knowledge., Comment: 10 pages, 9 figures, 1 tables, "Submitted to KDD 2025"

Black box attacks, where adversaries have limited knowledge of the target model, pose a significant threat to machine learning systems. Adversarial examples generated with a substitute model often suffer from limited transferability to the target model. While recent work explores ranking perturbations for improved success rates, these methods see only modest gains. We propose a novel strategy called PEAS that can boost the transferability of existing black box attacks. PEAS leverages the insight that samples which are perceptually equivalent exhibit significant variability in their adversarial transferability. Our approach first generates a set of images from an initial sample via subtle augmentations. We then evaluate the transferability of adversarial perturbations on these images using a set of substitute models. Finally, the most transferable adversarial example is selected and used for the attack. Our experiments show that PEAS can double the performance of existing attacks, achieving a 2.5x improvement in attack success rates on average over current ranking methods. We thoroughly evaluate PEAS on ImageNet and CIFAR-10, analyze hyperparameter impacts, and provide an ablation study to isolate each component's importance.

The increasing availability of public models begs the question: can we train neural networks that use other networks as input? Such models allow us to study different aspects of a given neural network, for example, determining the categories in a model's training dataset. However, machine learning on model weights is challenging as they often exhibit significant variation unrelated to the models' semantic properties (nuisance variation). Here, we identify a key property of real-world models: most public models belong to a small set of Model Trees, where all models within a tree are fine-tuned from a common ancestor (e.g., a foundation model). Importantly, we find that within each tree there is less nuisance variation between models. Concretely, while learning across Model Trees requires complex architectures, even a linear classifier trained on a single model layer often works within trees. While effective, these linear classifiers are computationally expensive, especially when dealing with larger models that have many parameters. To address this, we introduce Probing Experts (ProbeX), a theoretically motivated and lightweight method. Notably, ProbeX is the first probing method specifically designed to learn from the weights of a single hidden model layer. We demonstrate the effectiveness of ProbeX by predicting the categories in a model's training dataset based only on its weights. Excitingly, ProbeX can also map the weights of Stable Diffusion into a shared weight-language embedding space, enabling zero-shot model classification., Comment: Project page: https://horwitz.ai/probex/

We study the generalization properties of binary logistic classification in a simplified setting, for which a "memorizing" and "generalizing" solution can always be strictly defined, and elucidate empirically and analytically the mechanism underlying Grokking in its dynamics. We analyze the asymptotic long-time dynamics of logistic classification on a random feature model with a constant label and show that it exhibits Grokking, in the sense of delayed generalization and non-monotonic test loss. We find that Grokking is amplified when classification is applied to training sets which are on the verge of linear separability. Even though a perfect generalizing solution always exists, we prove the implicit bias of the logisitc loss will cause the model to overfit if the training data is linearly separable from the origin. For training sets that are not separable from the origin, the model will always generalize perfectly asymptotically, but overfitting may occur at early stages of training. Importantly, in the vicinity of the transition, that is, for training sets that are almost separable from the origin, the model may overfit for arbitrarily long times before generalizing. We gain more insights by examining a tractable one-dimensional toy model that quantitatively captures the key features of the full model. Finally, we highlight intriguing common properties of our findings with recent literature, suggesting that grokking generally occurs in proximity to the interpolation threshold, reminiscent of critical phenomena often observed in physical systems., Comment: 24 pages, 13 figures

Video detailed captioning is a key task which aims to generate comprehensive and coherent textual descriptions of video content, benefiting both video understanding and generation. In this paper, we propose AuroraCap, a video captioner based on a large multimodal model. We follow the simplest architecture design without additional parameters for temporal modeling. To address the overhead caused by lengthy video sequences, we implement the token merging strategy, reducing the number of input visual tokens. Surprisingly, we found that this strategy results in little performance loss. AuroraCap shows superior performance on various video and image captioning benchmarks, for example, obtaining a CIDEr of 88.9 on Flickr30k, beating GPT-4V (55.3) and Gemini-1.5 Pro (82.2). However, existing video caption benchmarks only include simple descriptions, consisting of a few dozen words, which limits research in this field. Therefore, we develop VDC, a video detailed captioning benchmark with over one thousand carefully annotated structured captions. In addition, we propose a new LLM-assisted metric VDCscore for bettering evaluation, which adopts a divide-and-conquer strategy to transform long caption evaluation into multiple short question-answer pairs. With the help of human Elo ranking, our experiments show that this benchmark better correlates with human judgments of video detailed captioning quality., Comment: Code, docs, weight, benchmark and training data are all avaliable at \href{https://rese1f.github.io/aurora-web/}{website}

Gene regulation is a dynamic process that underlies all aspects of human development, disease response, and other key biological processes. The reconstruction of temporal gene regulatory networks has conventionally relied on regression analysis, graphical models, or other types of relevance networks. With the large increase in time series single-cell data, new approaches are needed to address the unique scale and nature of this data for reconstructing such networks. Here, we develop a deep neural network, Marlene, to infer dynamic graphs from time series single-cell gene expression data. Marlene constructs directed gene networks using a self-attention mechanism where the weights evolve over time using recurrent units. By employing meta learning, the model is able to recover accurate temporal networks even for rare cell types. In addition, Marlene can identify gene interactions relevant to specific biological responses, including COVID-19 immune response, fibrosis, and aging., Comment: 10 pages, 5 figures

We describe an efficient algorithm to compute finite type invariants of type $k$ by first creating, for a given knot $K$ with $n$ crossings, a look-up table for all subdiagrams of $K$ of size $\lceil \frac{k}{2}\rceil$ indexed by dyadic intervals in $[0,2n-1]$. Using this algorithm, any such finite type invariant can be computed on an $n$-crossing knot in time $\sim n^{\lceil \frac{k}{2}\rceil}$, a lot faster than the previously best published bound of $\sim n^k$.

In his 1996 paper, Talagrand highlighted that the Law of Large Numbers (LLN) for independent random variables can be viewed as a geometric property of multidimensional product spaces. This phenomenon is known as the concentration of measure. To illustrate this profound connection between geometry and probability theory, we consider a seemingly intractable geometric problem in multidimensional Euclidean space and solve it using standard probabilistic tools such as the LLN and the Central Limit Theorem (CLT)., Comment: 14 pages, 3 figures

Recent years have seen significant growth of quantum technologies, and specifically quantum sensing, both in terms of the capabilities of advanced platforms and their applications. One of the leading platforms in this context is nitrogen-vacancy (NV) color centers in diamond, providing versatile, high-sensitivity, and high-resolution magnetic sensing. Nevertheless, current schemes for spin resonance magnetic sensing (as applied by NV quantum sensing) suffer from tradeoffs associated with sensitivity, dynamic range, and bandwidth. Here we address this issue, and implement machine learning tools to enhance NV magnetic sensing in terms of the sensitivity/bandwidth tradeoff in large dynamic range scenarios. We experimentally demonstrate this new approach, reaching an improvement in the relevant figure of merit by a factor of up to 5. Our results promote quantum machine learning protocols for sensing applications towards more feasible and efficient quantum technologies., Comment: 12 pages including supplementary, 6 figures

The study of coherence dynamics in open quantum systems, specifically addressing various physical realizations of quantum systems and environments, is a long-standing and central pillar of quantum science and technology. As such, a large body of work establishes a firm theoretical understanding of these processes. Nevertheless, a fundamental aspect of decoherence dynamics, namely the central limit theorem of qubit dephasing in the central spin model, which leads to a Gaussian approximation, lacks formal proof in realistically relevant scenarios. Here we prove this approximation for a bath depicted by an Ising spin system, in the presence of disorder and several (most relevant) functional forms of qubit-bath coupling. Importantly, we show that in certain cases, namely for short-range (exponentially decaying) coupling, this approximation breaks. These results provide a theoretical framework for studying decoherence dynamics in various systems and lead to insights into dephasing behavior with implications for applications in quantum information, quantum computing, and other quantum technologies.

Embodied vision-based real-world systems, such as mobile robots, require a careful balance between energy consumption, compute latency, and safety constraints to optimize operation across dynamic tasks and contexts. As local computation tends to be restricted, offloading the computation, ie, to a remote server, can save local resources while providing access to high-quality predictions from powerful and large models. However, the resulting communication and latency overhead has led to limited usability of cloud models in dynamic, safety-critical, real-time settings. To effectively address this trade-off, we introduce UniLCD, a novel hybrid inference framework for enabling flexible local-cloud collaboration. By efficiently optimizing a flexible routing module via reinforcement learning and a suitable multi-task objective, UniLCD is specifically designed to support the multiple constraints of safety-critical end-to-end mobile systems. We validate the proposed approach using a challenging, crowded navigation task requiring frequent and timely switching between local and cloud operations. UniLCD demonstrates improved overall performance and efficiency, by over 35% compared to state-of-the-art baselines based on various split computing and early exit strategies., Comment: ECCV 24

We consider the problem of estimating the direction-of-arrival (DoA) of a desired source located in a known region of interest in the presence of interfering sources and multipath. We propose an approach that precedes the DoA estimation and relies on generating a set of reference steering vectors. The steering vectors' generative model is a free space model, which is beneficial for many DoA estimation algorithms. The set of reference steering vectors is then used to compute a function that maps the received signals from the adverse environment to a reference domain free from interfering sources and multipath. We show theoretically and empirically that the proposed map, which is analogous to domain adaption, improves DoA estimation by mitigating interference and multipath effects. Specifically, we demonstrate a substantial improvement in accuracy when the proposed approach is applied before three commonly used beamformers: the delay-and-sum (DS), the minimum variance distortionless response (MVDR), and the Multiple Signal Classification (MUSIC).

We prove that a free profinite (pro-$p$) product over a set converging to 1 of countably many Demushkin groups of rank $\aleph_0$, $G_i$, that can be realized as absolute Galois groups, is isomorphic to an absolute Galois group if and only if $\log_pq(G_i)\to \infty$.

We present a novel approach for test-time adaptation via online self-training, consisting of two components. First, we introduce a statistical framework that detects distribution shifts in the classifier's entropy values obtained on a stream of unlabeled samples. Second, we devise an online adaptation mechanism that utilizes the evidence of distribution shifts captured by the detection tool to dynamically update the classifier's parameters. The resulting adaptation process drives the distribution of test entropy values obtained from the self-trained classifier to match those of the source domain, building invariance to distribution shifts. This approach departs from the conventional self-training method, which focuses on minimizing the classifier's entropy. Our approach combines concepts in betting martingales and online learning to form a detection tool capable of quickly reacting to distribution shifts. We then reveal a tight relation between our adaptation scheme and optimal transport, which forms the basis of our novel self-supervised loss. Experimental results demonstrate that our approach improves test-time accuracy under distribution shifts while maintaining accuracy and calibration in their absence, outperforming leading entropy minimization methods across various scenarios.

Topological deep learning (TDL) facilitates learning from data represented by topological structures. The primary model utilized in this setting is higher-order message-passing (HOMP), which extends traditional graph message-passing neural networks (MPNN) to diverse topological domains. Given the significant expressivity limitations of MPNNs, our paper aims to explore both the strengths and weaknesses of HOMP's expressive power and subsequently design novel architectures to address these limitations. We approach this from several perspectives: First, we demonstrate HOMP's inability to distinguish between topological objects based on fundamental topological and metric properties such as diameter, orientability, planarity, and homology. Second, we show HOMP's limitations in fully leveraging the topological structure of objects constructed using common lifting and pooling operators on graphs. Finally, we compare HOMP's expressive power to hypergraph networks, which are the most extensively studied TDL methods. We then develop two new classes of TDL models: multi-cellular networks (MCN) and scalable multi-cellular networks (SMCN). These models draw inspiration from expressive graph architectures. While MCN can reach full expressivity but is highly unscalable, SMCN offers a more scalable alternative that still mitigates many of HOMP's expressivity limitations. Finally, we construct a synthetic dataset, where TDL models are tasked with separating pairs of topological objects based on basic topological properties. We demonstrate that while HOMP is unable to distinguish between any of the pairs in the dataset, SMCN successfully distinguishes all pairs, empirically validating our theoretical findings. Our work opens a new design space and new opportunities for TDL, paving the way for more expressive and versatile models.

These lecture notes are derived from a graduate-level course in dynamic optimization, offering an introduction to techniques and models extensively used in management science, economics, operations research, engineering, and computer science. The course emphasizes the theoretical underpinnings of discrete-time dynamic programming models and advanced algorithmic strategies for solving these models. Unlike typical treatments, it provides a proof for the principle of optimality for upper semi-continuous dynamic programming, a middle ground between the simpler countable state space case \cite{bertsekas2012dynamic}, and the involved universally measurable case \cite{bertsekas1996stochastic}. This approach is sufficiently rigorous to include important examples such as dynamic pricing, consumption-savings, and inventory management models. The course also delves into the properties of value and policy functions, leveraging classical results \cite{topkis1998supermodularity} and recent developments. Additionally, it offers an introduction to reinforcement learning, including a formal proof of the convergence of Q-learning algorithms. Furthermore, the notes delve into policy gradient methods for the average reward case, presenting a convergence result for the tabular case in this context. This result is simple and similar to the discounted case but appears to be new.

Earthquake occurrence is notoriously difficult to predict. While some aspects of their spatiotemporal statistics can be relatively well captured by point-process models, very little is known regarding the magnitude of future events, and it is deeply debated whether it is possible to predict the magnitude of an earthquake before it starts. This is due both to the lack of information about fault conditions and to the inherent complexity of rupture dynamics. Consequently, even state of the art forecasting models typically assume no knowledge about the magnitude of future events besides the time-independent Gutenberg Richter (GR) distribution, which describes the marginal distribution over large regions and long times. This approach implicitly assumes that earthquake magnitudes are independent of previous seismicity and are identically distributed. In this work we challenge this view by showing that information about the magnitude of an upcoming earthquake can be directly extracted from the seismic history. We present MAGNET - MAGnitude Neural EsTimation model, an open-source, geophysically-inspired neural-network model for probabilistic forecasting of future magnitudes from cataloged properties: hypocenter locations, occurrence times and magnitudes of past earthquakes. Our history-dependent model outperforms stationary and quasi-stationary state of the art GR-based benchmarks, in real catalogs in Southern California, Japan and New-Zealand. This demonstrates that earthquake catalogs contain information about the magnitude of future earthquakes, prior to their occurrence. We conclude by proposing methods to apply the model in characterization of the preparatory phase of earthquakes, and in operational hazard alert and earthquake forecasting systems., Comment: 4 main figure, 1 main table

Large Language Models (LLMs) have demonstrated great performance in few-shot In-Context Learning (ICL) for a variety of generative and discriminative chemical design tasks. The newly expanded context windows of LLMs can further improve ICL capabilities for molecular inverse design and lead optimization. To take full advantage of these capabilities we developed a new semi-supervised learning method that overcomes the lack of experimental data available for many-shot ICL. Our approach involves iterative inclusion of LLM generated molecules with high predicted performance, along with experimental data. We further integrated our method in a multi-modal LLM which allows for the interactive modification of generated molecular structures using text instructions. As we show, the new method greatly improves upon existing ICL methods for molecular design while being accessible and easy to use for scientists.

We present a general analysis for the discovery potential of CP-violation (CPV) searches in scattering processes at TeV-scale colliders in an effective field theory framework, using the SMEFT basis for higher dimensional operators. In particular, we systematically examine the CP-violating sector of the SMEFT framework in some well motivated limiting cases, based on flavour symmetries of the underlying heavy theory. We show that, under naturality arguments of the underlying new physics (NP) and in the absence of (or suppressed) flavour-changing interactions, there is only a single operator, $Q_{t\phi} = \phi^\dagger \phi \left(\bar q_3 t \right) \tilde{\phi} $ which alters the top-Yukawa coupling, that can generate a non-vanishing CP-violating effect from tree-level SM$\times$NP interference terms. We find, however, that CPV from $Q_{t\phi} = \phi^\dagger \phi \left(\bar q_3 t \right) \tilde{\phi} $ is expected to be at best of $O(1\%)$ and, therefore, very challenging if at all measurable at the LHC or other future high-energy colliders. We then conclude that a potentially measurable CP-violating effect of $O(10\%)$ can arise in high-energy scattering processes ONLY if flavour-changing interactions are present in the underlying NP; in this case a sizable CPV can be generated at the tree-level by pure NP$\times$NP effects and not from SM$\times$NP interference. We provide several examples of CPV at the LHC and at a future $e^+e^-$ collider to support these statements., Comment: 15 pages, 5 figures, 6 tables

The channel output entropy of a transmitted sequence is the entropy of the possible channel outputs and similarly the channel input entropy of a received sequence is the entropy of all possible transmitted sequences. The goal of this work is to study these entropy values for the k-deletion, k-insertion channels, where exactly k symbols are deleted, inserted in the transmitted sequence, respectively. If all possible sequences are transmitted with the same probability then studying the input and output entropies is equivalent. For both the 1-deletion and 1-insertion channels, it is proved that among all sequences with a fixed number of runs, the input entropy is minimized for sequences with a skewed distribution of their run lengths and it is maximized for sequences with a balanced distribution of their run lengths. Among our results, we establish a conjecture by Atashpendar et al. which claims that for the 1-deletion channel, the input entropy is maximized by the alternating sequences over all binary sequences. This conjecture is also verified for the 2-deletion channel, where it is proved that constant sequences with a single run minimize the input entropy., Comment: arXiv admin note: substantial text overlap with arXiv:2202.03024

Enterprise chatbots, powered by generative AI, are emerging as key applications to enhance employee productivity. Retrieval Augmented Generation (RAG), Large Language Models (LLMs), and orchestration frameworks like Langchain and Llamaindex are crucial for building these chatbots. However, creating effective enterprise chatbots is challenging and requires meticulous RAG pipeline engineering. This includes fine-tuning embeddings and LLMs, extracting documents from vector databases, rephrasing queries, reranking results, designing prompts, honoring document access controls, providing concise responses, including references, safeguarding personal information, and building orchestration agents. We present a framework for building RAG-based chatbots based on our experience with three NVIDIA chatbots: for IT/HR benefits, financial earnings, and general content. Our contributions are three-fold: introducing the FACTS framework (Freshness, Architectures, Cost, Testing, Security), presenting fifteen RAG pipeline control points, and providing empirical results on accuracy-latency tradeoffs between large and small LLMs. To the best of our knowledge, this is the first paper of its kind that provides a holistic view of the factors as well as solutions for building secure enterprise-grade chatbots.", Comment: 8 pages, 6 figures, 2 tables, Preprint submission to ACM CIKM 2024

Blockchains implement monetary systems operated by committees of nodes. The robustness of established blockchains presents an opportunity to leverage their infrastructure for creating expansion chains. Expansion chains can provide additional functionality to the primary chain they leverage or implement separate functionalities, while benefiting from the primary chain's security and the stability of its tokens. Indeed, tools like Ethereum's EigenLayer enable nodes to stake (deposit collateral) on a primary chain to form a committee responsible for operating an expansion chain. But here is the rub. Classical protocols assume correct, well-behaved nodes stay correct indefinitely. Yet in our case, the stake incentivizes correctness--it will be slashed (revoked) if its owner deviates. Once a node withdraws its stake, there is no basis to assume its correctness. To address the new challenge, we present Aegis, an expansion chain based on primary-chain stake, assuming a bounded primary-chain write time. Aegis uses references from Aegis blocks to primary blocks to define committees, checkpoints on the primary chain to perpetuate decisions, and resets on the primary chain to establish a new committee if the previous one becomes obsolete. It ensures safety at all times and rapid progress when latency among Aegis nodes is low.

Metabolic models condense biochemical knowledge about organisms in a structured and standardised way. As large-scale network reconstructions are readily available for many organisms of interest, genome-scale models are being widely used among modellers and engineers. However, these large models can be difficult to analyse and visualise, and occasionally generate hard-to-interpret or even biologically unrealistic predictions. Out of the thousands of enzymatic reactions in a typical bacterial metabolism, only a few hundred comprise the metabolic pathways essential to produce energy carriers and biosynthetic precursors. These pathways carry relatively high flux, are central to maintaining and reproducing the cell, and provide precursors and energy to engineered metabolic pathways. Here, focusing on these central metabolic subsystems, we present a manually-curated medium-scale model of energy and biosynthesis metabolism for the well-studied prokaryote Escherichia coli K-12 MG1655. The model is a sub-network of the most recent genome-scale reconstruction, iML1515, and comes with an updated layer of database annotations, as well as a range of metabolic maps for visualisation. We enriched the stoichiometric network with extensive biological information and quantitative data, enhancing the scope and applicability of the model. In addition, here we assess the properties of this model in relation to its genome-scale parent and demonstrate the use of the network and supporting data in various scenarios, including enzyme-constrained flux balance analysis, elementary flux mode analysis, and thermodynamic analysis. Overall, we believe this model holds the potential to become a reference medium-scale metabolic model for E. coli., Comment: Minor stylistic changes. Extended discussion in Supplementary Information A.5

Recent improvements in generative AI made synthesizing fake images easy; as they can be used to cause harm, it is crucial to develop accurate techniques to identify them. This paper introduces "Locally Aware Deepfake Detection Algorithm" (LaDeDa), that accepts a single 9x9 image patch and outputs its deepfake score. The image deepfake score is the pooled score of its patches. With merely patch-level information, LaDeDa significantly improves over the state-of-the-art, achieving around 99% mAP on current benchmarks. Owing to the patch-level structure of LaDeDa, we hypothesize that the generation artifacts can be detected by a simple model. We therefore distill LaDeDa into Tiny-LaDeDa, a highly efficient model consisting of only 4 convolutional layers. Remarkably, Tiny-LaDeDa has 375x fewer FLOPs and is 10,000x more parameter-efficient than LaDeDa, allowing it to run efficiently on edge devices with a minor decrease in accuracy. These almost-perfect scores raise the question: is the task of deepfake detection close to being solved? Perhaps surprisingly, our investigation reveals that current training protocols prevent methods from generalizing to real-world deepfakes extracted from social media. To address this issue, we introduce WildRF, a new deepfake detection dataset curated from several popular social networks. Our method achieves the top performance of 93.7% mAP on WildRF, however the large gap from perfect accuracy shows that reliable real-world deepfake detection is still unsolved.

Subgraph Graph Neural Networks (Subgraph GNNs) enhance the expressivity of message-passing GNNs by representing graphs as sets of subgraphs. They have shown impressive performance on several tasks, but their complexity limits applications to larger graphs. Previous approaches suggested processing only subsets of subgraphs, selected either randomly or via learnable sampling. However, they make suboptimal subgraph selections or can only cope with very small subset sizes, inevitably incurring performance degradation. This paper introduces a new Subgraph GNNs framework to address these issues. We employ a graph coarsening function to cluster nodes into super-nodes with induced connectivity. The product between the coarsened and the original graph reveals an implicit structure whereby subgraphs are associated with specific sets of nodes. By running generalized message-passing on such graph product, our method effectively implements an efficient, yet powerful Subgraph GNN. Controlling the coarsening function enables meaningful selection of any number of subgraphs while, contrary to previous methods, being fully compatible with standard training techniques. Notably, we discover that the resulting node feature tensor exhibits new, unexplored permutation symmetries. We leverage this structure, characterize the associated linear equivariant layers and incorporate them into the layers of our Subgraph GNN architecture. Extensive experiments on multiple graph learning benchmarks demonstrate that our method is significantly more flexible than previous approaches, as it can seamlessly handle any number of subgraphs, while consistently outperforming baseline approaches., Comment: Preprint, under review

Simultaneous speech-to-speech translation (S2ST) holds the promise of breaking down communication barriers and enabling fluid conversations across languages. However, achieving accurate, real-time translation through mobile devices remains a major challenge. We introduce SimulTron, a novel S2ST architecture designed to tackle this task. SimulTron is a lightweight direct S2ST model that uses the strengths of the Translatotron framework while incorporating key modifications for streaming operation, and an adjustable fixed delay. Our experiments show that SimulTron surpasses Translatotron 2 in offline evaluations. Furthermore, real-time evaluations reveal that SimulTron improves upon the performance achieved by Translatotron 1. Additionally, SimulTron achieves superior BLEU scores and latency compared to previous real-time S2ST method on the MuST-C dataset. Significantly, we have successfully deployed SimulTron on a Pixel 7 Pro device, show its potential for simultaneous S2ST on-device.