Search

We address the problem of gaze target estimation, which aims to predict where a person is looking in a scene. Predicting a person's gaze target requires reasoning both about the person's appearance and the contents of the scene. Prior works have developed increasingly complex, hand-crafted pipelines for gaze target estimation that carefully fuse features from separate scene encoders, head encoders, and auxiliary models for signals like depth and pose. Motivated by the success of general-purpose feature extractors on a variety of visual tasks, we propose Gaze-LLE, a novel transformer framework that streamlines gaze target estimation by leveraging features from a frozen DINOv2 encoder. We extract a single feature representation for the scene, and apply a person-specific positional prompt to decode gaze with a lightweight module. We demonstrate state-of-the-art performance across several gaze benchmarks and provide extensive analysis to validate our design choices. Our code is available at: http://github.com/fkryan/gazelle .

Nitrogen-vacancy (NV) centers in diamonds have been an epicenter of research for diverse applications in quantum technologies. It is therefore imperative that their fabrication techniques are well understood and characterized for the technological scalability of these applications. A comparative study of the optical and spin properties of NVs created by ion implantation and chemical vapor deposition delta-doping is thus presented, combined with an investigation on the impact of annealing in vacuum at different temperatures. In addition, nanopillars are fabricated with electron beam lithography and reactive ion etching for enhanced photon collection efficiency. An extensive combination of characterization techniques is employed. Notably, the smallest nanopillars present fluorescence enhancements of factor around 50, compared to the unstructured regions. Annealing is also demonstrated to increase the optical contrast between the NVs' electronic states, the coherence and relaxation times both in bulk as in pillars. Regarding the NV preparation technique, the delta-doping is shown to create NVs with less lattice defects and strain compared to implantation., Comment: 11 pages, 10 figures

We analyzed the 7.92$\times 10^{11}$ cosmic-ray-induced muon events collected by the IceCube Neutrino Observatory from May 13, 2011, when the fully constructed experiment started to take data, to May 12, 2023. This dataset provides an up-to-date cosmic-ray arrival direction distribution in the Southern Hemisphere with unprecedented statistical accuracy covering more than a full period length of a solar cycle. Improvements in Monte Carlo event simulation and better handling of year-to-year differences in data processing significantly reduce systematic uncertainties below the level of statistical fluctuations compared to the previously published results. We confirm the observation of a change in the angular structure of the cosmic-ray anisotropy between 10 TeV and 1 PeV, more specifically in the 100-300 TeV energy range.

Since the initial discovery of two-dimensional van der Waals (vdW) materials, significant effort has been made to incorporate the three properties of magnetism, band structure topology, and strong electron correlations $-$ to leverage emergent quantum phenomena and expand their potential applications. However, the discovery of a single vdW material that intrinsically hosts all three ingredients has remained an outstanding challenge. Here we report the discovery of a Kondo-interacting topological antiferromagnet in the vdW 5$f$ electron system UOTe. It has a high antiferromagnetic (AFM) transition temperature of 150 K, with a unique AFM configuration that breaks the combined parity and time reversal ($PT$) symmetry in an even number of layers while maintaining zero net magnetic moment. Our angle-resolved photoemission spectroscopy (ARPES) measurements reveal Dirac bands near the Fermi level, which combined with our theoretical calculations demonstrate UOTe as an AFM Dirac semimetal. Within the AFM order, we observed the presence of the Kondo interaction, as evidenced by the emergence of a 5$f$ flat band near the Fermi level below 100 K and hybridization between the Kondo band and the Dirac band. Our density functional theory calculations in its bilayer form predict UOTe as a rare example of a fully-compensated AFM Chern insulator., Comment: Accepted to Advanced Materials

The Activated Random Walk (ARW) model is a promising candidate for demonstrating self-organized criticality due to its potential for universality. Recent studies have shown that the ARW model exhibits a well-defined critical density in one dimension, supporting its universality. In this paper, we extend these results by demonstrating that the ARW model on $\mathbb{Z}$, with a single initially active particle and all other particles sleeping, maintains the same critical density. Our findings relax the previous assumption that required all particles to be initially active. This provides further evidence of the ARW model's robustness and universality in depicting self-organized criticality.

Text-to-image diffusion models have impactful applications in art, design, and entertainment, yet these technologies also pose significant risks by enabling the creation and dissemination of misinformation. Although recent advancements have produced AI-generated image detectors that claim robustness against various augmentations, their true effectiveness remains uncertain. Do these detectors reliably identify images with different levels of augmentation? Are they biased toward specific scenes or data distributions? To investigate, we introduce SEMI-TRUTHS, featuring 27,600 real images, 223,400 masks, and 1,472,700 AI-augmented images that feature targeted and localized perturbations produced using diverse augmentation techniques, diffusion models, and data distributions. Each augmented image is accompanied by metadata for standardized and targeted evaluation of detector robustness. Our findings suggest that state-of-the-art detectors exhibit varying sensitivities to the types and degrees of perturbations, data distributions, and augmentation methods used, offering new insights into their performance and limitations. The code for the augmentation and evaluation pipeline is available at https://github.com/J-Kruk/SemiTruths., Comment: Accepted at NeurIPS 2024 Track Datasets & Benchmarks Track

Axions are a well-motivated candidate for dark matter. The preeminent method to search for axion dark matter is known as the axion haloscope, which makes use of the conversion of axions to photons in a large magnetic field. Due to the weak coupling of axions to photons however, the expected signal strength is exceptionally small. To increase signal strength, many haloscopes make use of resonant enhancement and high gain amplifiers, while also taking measures to keep receiver noise as low as possible such as the use of dilution refrigerators and ultra low-noise electronics. In this paper we derive the theoretical noise model based on the sources of noise found within a typical axion haloscope receiver chain, using the Axion Dark Matter eXperiment (ADMX) as a case study. We present examples of different noise calibration measurements at 1280~MHz using a variable temperature stage with ADMX during its most recent data taking run. The consistency between the measurements and the detailed model provide suggestions for future improvements within ADMX and other axion haloscopes to reach a lower noise temperature and to simplify the receiver chain design.

Today, cheap numerical hardware offers huge amounts of parallel computing power, much of which is used for the task of fitting neural networks to data. Adoption of this hardware to accelerate statistical Markov chain Monte Carlo (MCMC) applications has been much slower. In this chapter, we suggest some patterns for speeding up MCMC workloads using the hardware (e.g., GPUs, TPUs) and software (e.g., PyTorch, JAX) that have driven progress in deep learning over the last fifteen years or so. We offer some intuitions for why these new systems are so well suited to MCMC, and show some examples (with code) where we use them to achieve dramatic speedups over a CPU-based workflow. Finally, we discuss some potential pitfalls to watch out for.

Bak, Tang, and Wiesenfeld introduced self-organized criticality with the example of a growing sandpile that reaches then sustains a critical density. They presented the abelian sandpile as a potential model. About a decade later, Dickman, Mu\~noz, Vespignani, and Zapperi claimed that self-organizing sandpile models share their critical densities with parametrized model versions that have conventional phase transitions. Fey, Levine, and Wilson subsequently demonstrated that this fails to occur with the abelian sandpile model. However, the primary subject of Dickman et al.'s work was stochastic sandpile models, and it was widely speculated that their paradigm might still hold in that setting. To this end, we confirm a conjecture of Levine and Silvestri by providing the first rigorous proof that activated random walk in dimension one is driven directly to a critical density corresponding to a parametrized phase transition and then remains there., Comment: 12 pages

Stretchable electronics capable of conforming to nonplanar and dynamic human body surfaces are central for creating implantable and on-skin devices for high-fidelity monitoring of diverse physiological signals. While various strategies have been developed to produce stretchable devices, the signals collected from such devices are often highly sensitive to local strain, resulting in inevitable convolution with surface strain-induced motion artifacts that are difficult to distinguish from intrinsic physiological signals. Here we report all-printed super stretchable strain-insensitive bioelectronics using a unique universal gradient interface (UGI) to bridge the gap between soft biomaterials and stiff electronic materials. Leveraging a versatile aerosol-based multi-materials printing technique that allows precise spatial control over the local stiffnesses with submicron resolution, the UGI enables strain-insensitive electronic devices with negligible resistivity changes under a 180% stretch ratio. We demonstrate various stretchable devices directly printed on the UGI for on-skin health monitoring with high signal quality and near perfect immunity to motion artifacts, including semiconductor-based photodetectors for sensing blood oxygen saturation levels and metal-based temperature sensors. The concept in this work will significantly simplify the fabrication and accelerate the development of a broad range of wearable and implantable bioelectronics for real-time health monitoring and personalized therapeutics.

The scale and diversity of demonstration data required for imitation learning is a significant challenge. We present EgoMimic, a full-stack framework which scales manipulation via human embodiment data, specifically egocentric human videos paired with 3D hand tracking. EgoMimic achieves this through: (1) a system to capture human embodiment data using the ergonomic Project Aria glasses, (2) a low-cost bimanual manipulator that minimizes the kinematic gap to human data, (3) cross-domain data alignment techniques, and (4) an imitation learning architecture that co-trains on human and robot data. Compared to prior works that only extract high-level intent from human videos, our approach treats human and robot data equally as embodied demonstration data and learns a unified policy from both data sources. EgoMimic achieves significant improvement on a diverse set of long-horizon, single-arm and bimanual manipulation tasks over state-of-the-art imitation learning methods and enables generalization to entirely new scenes. Finally, we show a favorable scaling trend for EgoMimic, where adding 1 hour of additional hand data is significantly more valuable than 1 hour of additional robot data. Videos and additional information can be found at https://egomimic.github.io/

Understanding the genetic basis of complex traits is a longstanding challenge in the field of genomics. Genome-wide association studies (GWAS) have identified thousands of variant-trait associations, but most of these variants are located in non-coding regions, making the link to biological function elusive. While traditional approaches, such as transcriptome-wide association studies (TWAS), have advanced our understanding by linking genetic variants to gene expression, they often overlook gene-gene interactions. Here, we review current approaches to integrate different molecular data, leveraging machine learning methods to identify gene modules based on co-expression and functional relationships. These integrative approaches, like PhenoPLIER, combine TWAS and drug-induced transcriptional profiles to effectively capture biologically meaningful gene networks. This integration provides a context-specific understanding of disease processes while highlighting both core and peripheral genes. These insights pave the way for novel therapeutic targets and enhance the interpretability of genetic studies in personalized medicine., Comment: 26 pages, 4 figures GitHub: https://github.com/pivlab/annual_review_of_biomedical_data_science HTML: https://pivlab.github.io/annual_review_of_biomedical_data_science/

Reliability is a fundamental challenge in operating large-scale machine learning (ML) infrastructures, particularly as the scale of ML models and training clusters continues to grow. Despite decades of research on infrastructure failures, the impact of job failures across different scales remains unclear. This paper presents a view of managing two large, multi-tenant ML clusters, providing quantitative analysis, operational experience, and our own perspective in understanding and addressing reliability concerns at scale. Our analysis reveals that while large jobs are most vulnerable to failures, smaller jobs make up the majority of jobs in the clusters and should be incorporated into optimization objectives. We identify key workload properties, compare them across clusters, and demonstrate essential reliability requirements for pushing the boundaries of ML training at scale. We hereby introduce a taxonomy of failures and key reliability metrics, analyze 11 months of data from two state-of-the-art ML environments with over 150 million A100 GPU hours and 4 million jobs. Building on our data, we fit a failure model to project Mean Time to Failure for various GPU scales. We further propose a method to estimate a related metric, Effective Training Time Ratio, as a function of job parameters, and we use this model to gauge the efficacy of potential software mitigations at scale. Our work provides valuable insights and future research directions for improving the reliability of AI supercomputer clusters, emphasizing the need for flexible, workload-agnostic, and reliability-aware infrastructure, system software, and algorithms.

Recent model merging methods demonstrate that the parameters of fully-finetuned models specializing in distinct tasks can be combined into one model capable of solving all tasks without retraining. Yet, this success does not transfer well when merging LoRA finetuned models. We study this phenomenon and observe that the weights of LoRA finetuned models showcase a lower degree of alignment compared to their fully-finetuned counterparts. We hypothesize that improving this alignment is key to obtaining better LoRA model merges, and propose KnOTS to address this problem. KnOTS uses the SVD to jointly transform the weights of different LoRA models into an aligned space, where existing merging methods can be applied. In addition, we introduce a new benchmark that explicitly evaluates whether merged models are general models. Notably, KnOTS consistently improves LoRA merging by up to 4.3% across several vision and language benchmarks, including our new setting. We release our code at: https://github.com/gstoica27/KnOTS.

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.

We provide new characterizations of the $BMO$-Sobolev space $I_{\alpha}(BMO)$ for the range $0 < \alpha <2$. When $0 < \alpha <1$, our characterizations are in terms of square functions measuring multiscale approximation of constants, and when $1 \leq \alpha <2$ our characterizations are in terms of square functions measuring multiscale approximation by linear functions., Comment: 26 pages

Summary: ipd is an open-source R software package for the downstream modeling of an outcome and its associated features where a potentially sizable portion of the outcome data has been imputed by an artificial intelligence or machine learning (AI/ML) prediction algorithm. The package implements several recent proposed methods for inference on predicted data (IPD) with a single, user-friendly wrapper function, ipd. The package also provides custom print, summary, tidy, glance, and augment methods to facilitate easy model inspection. This document introduces the ipd software package and provides a demonstration of its basic usage. Availability: ipd is freely available on CRAN or as a developer version at our GitHub page: github.com/ipd-tools/ipd. Full documentation, including detailed instructions and a usage `vignette' are available at github.com/ipd-tools/ipd. Contact: jtleek@fredhutch.org and tylermc@uw.edu, Comment: 5 pages, 1 figure

The Axion Dark Matter eXperiment is sensitive to narrow axion flows, given axions compose a fraction of the dark matter with a non-negligible local density. Detecting these low-velocity dispersion flows requires a high spectral resolution and careful attention to the expected signal modulation due to Earth's motion. We report an exclusion on the local axion dark matter density in narrow flows of $\rho_a \gtrsim 0.03\,\mathrm{GeV/cm^3}$ and $\rho_a \gtrsim 0.004\,\mathrm{GeV/cm^3}$ for Dine-Fischler-Srednicki-Zhitnitski and Kim-Shifman-Vainshtein-Zakharov axion-photon couplings, respectively, over the mass range $3.3-4.2\,\mu\text{eV}$. Measurements were made at selected resolving powers to allow for a range of possible velocity dispersions., Comment: 7 pages, 3 figures

We present multi-epoch optical spectropolarimetric and imaging polarimetric observations of the nearby Type II supernova (SN) 2023ixf discovered in M101 at a distance of 6.85 Mpc. The first imaging polarimetric observations were taken +2.33 days (60085.08 MJD) after the explosion, while the last imaging polarimetric data points (+73.19 and +76.19 days) were acquired after the fall from the light curve plateau. At +2.33 days there is strong evidence of circumstellar material (CSM) interaction in the spectra and the light curve. A significant level of polarization $P_r = 0.88\pm 0.06 \% $ seen during this phase indicates that this CSM is aspherical. We find that the polarization evolves with time toward the interstellar polarization level ($0.35\%$) during the photospheric phase, which suggests that the recombination photosphere is spherically symmetric. There is a jump in polarization ($P_r =0.65 \pm 0.08 \% $) at +73.19 days when the light curve falls from the plateau. This is a phase where polarimetric data is sensitive to non-spherical inner ejecta or a decrease in optical depth into the single scattering regime. We also present spectropolarimetric data that reveal line (de)polarization during most of the observed epochs. In addition, at +14.50 days we see an "inverse P Cygn" profile in the H and He line polarization, which clearly indicates the presence of asymmetrically distributed material overlying the photosphere. The overall temporal evolution of polarization is typical for Type II SNe, but the high level of polarization during the rising phase has only been observed in SN 2023ixf., Comment: 14 pages, 7 figures, submitted to ApJL, comments welcome

This white paper is the result of a collaboration by those that attended a workshop at the Facility for Rare Isotope Beams (FRIB), organized by the FRIB Theory Alliance (FRIB-TA), on Theoretical Justifications and Motivations for Early High-Profile FRIB Experiments. It covers a wide range of topics related to the science that will be explored at FRIB. After a brief introduction, the sections address: (II) Overview of theoretical methods, (III) Experimental capabilities, (IV) Structure, (V) Near-threshold Physics, (VI) Reaction mechanisms, (VII) Nuclear equations of state, (VIII) Nuclear astrophysics, (IX) Fundamental symmetries, and (X) Experimental design and uncertainty quantification., Comment: 227 pages, 24 figures

Animal vocalization denoising is a task similar to human speech enhancement, a well-studied field of research. In contrast to the latter, it is applied to a higher diversity of sound production mechanisms and recording environments, and this higher diversity is a challenge for existing models. Adding to the challenge and in contrast to speech, we lack large and diverse datasets comprising clean vocalizations. As a solution we use as training data pseudo-clean targets, i.e. pre-denoised vocalizations, and segments of background noise without a vocalization. We propose a train set derived from bioacoustics datasets and repositories representing diverse species, acoustic environments, geographic regions. Additionally, we introduce a non-overlapping benchmark set comprising clean vocalizations from different taxa and noise samples. We show that that denoising models (demucs, CleanUNet) trained on pseudo-clean targets obtained with speech enhancement models achieve competitive results on the benchmarking set. We publish data, code, libraries, and demos https://mariusmiron.com/research/biodenoising., Comment: 5 pages, 2 tables

The Building Equitable Learning Environments (BELE) network is a diverse collaborative of organizations working together to advance equity in education. The BELE Network research and learning team, and are responsible for documenting the network learnings, which are grounded in seven "Essential Actions" and reflected in shared annual aims. This brief shares learnings from BELE district partnerships around the fifth Essential Action: invest in staff. To support this Essential Action, Collaborative for Academic, Social, and Emotional Learning (CASEL) developed the Transformative (tSEL) Adult Learning Series, a set of six interactive workshops that build an opportunity for adults themselves to deepen their self-awareness and discover how they can curate equitable learning environments for young people by leveraging the five focal constructs (identity, agency, belonging, collaborative problem-solving, and curiosity). Through this series, education leaders are making investments for in-service professional learning that supports educator well-being and promotes equitable learning environments for adults and young people. Investing in staff actualizes the value of continuous improvement, which fosters a culture of expansion, growth, and development among educators.

We introduce MCUBench, a benchmark featuring over 100 YOLO-based object detection models evaluated on the VOC dataset across seven different MCUs. This benchmark provides detailed data on average precision, latency, RAM, and Flash usage for various input resolutions and YOLO-based one-stage detectors. By conducting a controlled comparison with a fixed training pipeline, we collect comprehensive performance metrics. Our Pareto-optimal analysis shows that integrating modern detection heads and training techniques allows various YOLO architectures, including legacy models like YOLOv3, to achieve a highly efficient tradeoff between mean Average Precision (mAP) and latency. MCUBench serves as a valuable tool for benchmarking the MCU performance of contemporary object detectors and aids in model selection based on specific constraints., Comment: Code and data are available at https://github.com/Deeplite/deeplite-torch-zoo

Reconfigurable Intelligent Surfaces (RISs) can be designed in various ways. A previously proposed semidefinite relaxation-based optimization method for maximizing power transfer efficiency showed promise, but earlier results were only theoretical. This paper evaluates a small RIS at 3.55GHz, the center of the 5G band "n78", for practical verification of this method. The presented results not only empirically confirm the desired performance of the optimized RIS, but also affirm the optimality of the resulting reactance values. Additionally, this paper discusses several practical aspects of RIS design and measurement, such as the operation of varactor diodes and time gating to omit the direct line-of-sight (LOS) path., Comment: 5 pages, conference

Physical Unclonable Functions (PUFs) leverage signal variations that occur within the device as a source of entropy. On-chip instrumentation is utilized by some PUF architectures to measure and digitize these variations, which are then processed into bitstrings and secret keys for use in security functions such as authentication and encryption. In many cases, the variations in the measured signals are introduced by a sequence of components in the circuit structure defined by the PUF architecture. In particular, the Hardware-Embedded deLay PUF (HELP) measures delay variations that occur in combinational logic paths on Field Programmable Gate Arrays (FPGAs), which are composed of a set of interconnecting wires (nodes) and look-up tables (LUTs). Previous investigations of variations in these path delays show that it is possible to derive high quality bitstrings, i.e., those which exhibit high levels of uniqueness and randomness across the device population. However, the underlying source and level of variations associated with the constituent components of the paths remain unknown. In this paper, we apply statistical averaging and differencing techniques to derive estimates for the delay variation associated with an FPGA's basic components, namely LUTs and nodes, as a means of fully characterizing the PUF's source of entropy. The analysis is carried out on a set of 50,015 path delay measurements collected from a set of 20 Xilinx Zynq 7020 SoC-class FPGAs, on which 25 identical instances of a functional unit are instantiated, for a total of 500 instances.

We report the results of a QCD axion dark matter search with discovery ability for Dine-Fischler-Srednicki-Zhitnitsky (DFSZ) axions using an axion haloscope. Sub-Kelvin noise temperatures are reached with an ultra low-noise Josephson parametric amplifier cooled by a dilution refrigerator. This work excludes (with a 90% confidence level) DFSZ axions with masses between 3.27 to 3.34 {\mu}eV, assuming a standard halo model with a local energy density of 0.45 GeV/cm${}^3$ made up 100% of axions.

This survey examines the broad suite of methods and models for combining machine learning with physics knowledge for prediction and forecast, with a focus on partial differential equations. These methods have attracted significant interest due to their potential impact on advancing scientific research and industrial practices by improving predictive models with small- or large-scale datasets and expressive predictive models with useful inductive biases. The survey has two parts. The first considers incorporating physics knowledge on an architectural level through objective functions, structured predictive models, and data augmentation. The second considers data as physics knowledge, which motivates looking at multi-task, meta, and contextual learning as an alternative approach to incorporating physics knowledge in a data-driven fashion. Finally, we also provide an industrial perspective on the application of these methods and a survey of the open-source ecosystem for physics-informed machine learning., Comment: 56 pages, 8 figures, 2 tables

This is a report on recent work, with Wen-Ching Winnie Li and Ling Long. In that work explicit formulas are given, involving hypergeometric character sums, for the traces of Hecke operators $T_p$ acting spaces of cusp forms $S_k(\Gamma)$ of weight $k$ for certain arithmetically defined Fuchsian subgroups $\Gamma\subset \mathrm{SL}_2 (\mathbf{R})$. In particular we consider the groups attached to the quaternion division algebra $B_6$ over $\mathbf{Q}$ of discriminant 6., Comment: An overview of the background of the joint paper with Wen-Ching W. Li and Ling Long on "Traces of Hecke Operators via Hypergeometric Character Sums" (arXiv: 2408.02918)

Entropy or randomness represents a foundational security property in security-related operations, such as key generation. Key generation in turn is central to security protocols such as authentication and encryption. Physical unclonable functions (PUF) are hardware-based primitives that can serve as key generation engines in modern microelectronic devices and applications. PUFs derive entropy from manufacturing variations that exist naturally within and across otherwise identical copies of a device. However, the levels of random variations that represent entropy, which are strongly correlated to the quality of the PUF-generated bitstrings, vary from one manufacturer to another. In this paper, we evaluate entropy across a set of devices manufactured by three mainstream FPGA vendors, Xilinx, Altera and Microsemi. The devices selected for evaluation are considered low-end commercial devices to make the analysis relevant to IoT applications. The SiRF PUF is used in the evaluation, and is constructed nearly identically across the three vendor devices, setting aside minor differences that exist in certain logic element primitives used within the PUF architecture, and which have only a minor impact on our comparative analysis. The SiRF PUF uses a high-resolution time-to-digital converter (TDC) crafted from high-speed carry-chain logic embedded within each device to measure path delays in an engineered netlist of logic gates as a source of entropy. Therefore, our analysis includes an evaluation of actual path delay variation as it exists across the three device classes, as well as a statistical evaluation of the PUF-generated bitstrings. A reliablity analysis is also provided using data collected in industrial-standard temperature experiments to round out the evaluation of important statistical properties of the PUF., Comment: 15 pages, 22 figures, IEEE journal

In this paper we obtain explicit formulas for the traces of Hecke operators on spaces of cusp forms in certain instances related to arithmetic triangle groups. These expressions are in terms of hypergeometric character sums over finite fields, a theory developed largely by Greene, Katz, Beukers-Cohen-Mellit, and Fuselier-Long-Ramakrishna-Swisher-Tu. Our approach, in contrast to the previous works, is uniform and more geometric, and it works equally well for forms on elliptic modular curves and Shimura curves. The same method can be applied to obtain eigenvalues of Hecke operators as well.

In this work, we introduce Gemma 2, a new addition to the Gemma family of lightweight, state-of-the-art open models, ranging in scale from 2 billion to 27 billion parameters. In this new version, we apply several known technical modifications to the Transformer architecture, such as interleaving local-global attentions (Beltagy et al., 2020a) and group-query attention (Ainslie et al., 2023). We also train the 2B and 9B models with knowledge distillation (Hinton et al., 2015) instead of next token prediction. The resulting models deliver the best performance for their size, and even offer competitive alternatives to models that are 2-3 times bigger. We release all our models to the community.

Reinforcement learning from human feedback (RLHF) is a key driver of quality and safety in state-of-the-art large language models. Yet, a surprisingly simple and strong inference-time strategy is Best-of-N sampling that selects the best generation among N candidates. In this paper, we propose Best-of-N Distillation (BOND), a novel RLHF algorithm that seeks to emulate Best-of-N but without its significant computational overhead at inference time. Specifically, BOND is a distribution matching algorithm that forces the distribution of generations from the policy to get closer to the Best-of-N distribution. We use the Jeffreys divergence (a linear combination of forward and backward KL) to balance between mode-covering and mode-seeking behavior, and derive an iterative formulation that utilizes a moving anchor for efficiency. We demonstrate the effectiveness of our approach and several design choices through experiments on abstractive summarization and Gemma models. Aligning Gemma policies with BOND outperforms other RLHF algorithms by improving results on several benchmarks.

Pediatric central nervous system tumors are the leading cause of cancer-related deaths in children. The five-year survival rate for high-grade glioma in children is less than 20%. The development of new treatments is dependent upon multi-institutional collaborative clinical trials requiring reproducible and accurate centralized response assessment. We present the results of the BraTS-PEDs 2023 challenge, the first Brain Tumor Segmentation (BraTS) challenge focused on pediatric brain tumors. This challenge utilized data acquired from multiple international consortia dedicated to pediatric neuro-oncology and clinical trials. BraTS-PEDs 2023 aimed to evaluate volumetric segmentation algorithms for pediatric brain gliomas from magnetic resonance imaging using standardized quantitative performance evaluation metrics employed across the BraTS 2023 challenges. The top-performing AI approaches for pediatric tumor analysis included ensembles of nnU-Net and Swin UNETR, Auto3DSeg, or nnU-Net with a self-supervised framework. The BraTSPEDs 2023 challenge fostered collaboration between clinicians (neuro-oncologists, neuroradiologists) and AI/imaging scientists, promoting faster data sharing and the development of automated volumetric analysis techniques. These advancements could significantly benefit clinical trials and improve the care of children with brain tumors.

Purpose: Single-neutron adding data was collected in order to determine the distribution of the single-neutron strength of the $0f_{7/2}$, $1p_{3/2}$, $1p_{1/2}$ and $0f_{5/2}$ orbitals outside of $Z=16, N=18$, $^{34}$S. Methods: The $^{34}$S($d$,$p$)$^{35}$S reaction has been measured at 8 MeV/u to investigate cross sections to excited states in $^{35}$S. Outgoing proton yields and momenta were analyzed by the Super-Enge Split-Pole Spectrograph in conjunction with the CeBrA demonstrator located at the John D. Fox Laboratory at Florida State University. Angular distributions were compared with Distorted Wave Born Approximation calculations in order to extract single-neutron spectroscopic overlaps. Results: Spectroscopic overlaps and strengths were determined for states in $^{35}$S up through 6 MeV in excitation energy. Each orbital was observed to have fragmented strength where a single level carried the majority. The single-neutron centroids of the $0f_{7/2}$, $1p_{3/2}$, $1p_{1/2}$ and $0f_{5/2}$ orbitals were determined to be $2360^{+90}_{-40}$ keV, $3280^{+80}_{-50}$ keV, $4780^{+60}_{-40}$ keV, and $\gtrsim7500$ keV, respectively. Conclusion: A previous discrepancy in the literature with respect to distribution of the neutron $1p_{1/2}$ strength was resolved. The integration of the normalized spectroscopic strengths, up to 5.1 MeV in excitation energy, revealed fully-vacant occupancies for the $0f_{7/2}$, $1p_{3/2}$, and $1p_{1/2}$ orbitals, as expected. The spacing in the single-neutron energies highlighted a reduction in the traditional $N=28$ shell-gap, relative to both the $1p$ spin-orbit energy difference ($N=32$) and the lower limit on the $N=34$ shell spacing., Comment: Updated draft