Search

Teaching performance assessments (TPAs) can be used to assess the readiness of potential teachers because they require candidates to provide evidence of their planning and teaching skills through classroom videos accompanied by commentaries Ensuring that teaching candidates are well prepared to enter the classroom is a critical mission for teacher preparation programs and state agencies that approve programs and set teacher licensure standards. Teaching performance assessments (TPAs) can be used to assess the readiness of potential teachers because they require candidates to provide evidence of their teaching knowledge and skills through classroom videos, lesson plans, student work, and analysis of teaching and learning. Multiple studies have found that TPA scores predict effectiveness once candidates enter the classroom as licensed teachers. TPAs have been adopted in at least 16 states as a requirement of either teacher preparation program completion or initial licensure. California, the focus of this study, was one of the first states to adopt a TPA as a licensure requirement for beginning teachers. The report on which this brief is based explored whether certain preparation experiences predicted success on the California Teaching Performance Assessment (CalTPA) or the educative Teaching Performance Assessment (edTPA), the two widely available TPAs used across California preparation programs. Focusing on the 2021-22 and 2022-23 school years, this analysis examined whether TPA success for more than 18,000 California teaching candidates varied by preparation pathway, program, and the nature of preparation experiences as reported by respondents on the annual survey administered to all those completing preparation and applying for their preliminary teaching credential with the California Commission on Teacher Credentialing (CTC).

We introduce a novel method for human shape and pose recovery that can fully leverage multiple static views. We target fixed-multiview people monitoring, including elderly care and safety monitoring, in which calibrated cameras can be installed at the corners of a room or an open space but whose configuration may vary depending on the environment. Our key idea is to formulate it as neural optimization. We achieve this with HeatFormer, a neural optimizer that iteratively refines the SMPL parameters given multiview images, which is fundamentally agonistic to the configuration of views. HeatFormer realizes this SMPL parameter estimation as heat map generation and alignment with a novel transformer encoder and decoder. We demonstrate the effectiveness of HeatFormer including its accuracy, robustness to occlusion, and generalizability through an extensive set of experiments. We believe HeatFormer can serve a key role in passive human behavior modeling.

This paper introduces a novel clothed human model that can be learned from multiview RGB videos, with a particular emphasis on recovering physically accurate body and cloth movements. Our method, Position Based Dynamic Gaussians (PBDyG), realizes ``movement-dependent'' cloth deformation via physical simulation, rather than merely relying on ``pose-dependent'' rigid transformations. We model the clothed human holistically but with two distinct physical entities in contact: clothing modeled as 3D Gaussians, which are attached to a skinned SMPL body that follows the movement of the person in the input videos. The articulation of the SMPL body also drives physically-based simulation of the clothes' Gaussians to transform the avatar to novel poses. In order to run position based dynamics simulation, physical properties including mass and material stiffness are estimated from the RGB videos through Dynamic 3D Gaussian Splatting. Experiments demonstrate that our method not only accurately reproduces appearance but also enables the reconstruction of avatars wearing highly deformable garments, such as skirts or coats, which have been challenging to reconstruct using existing methods.

We introduce the Aya Expanse model family, a new generation of 8B and 32B parameter multilingual language models, aiming to address the critical challenge of developing highly performant multilingual models that match or surpass the capabilities of monolingual models. By leveraging several years of research at Cohere For AI and Cohere, including advancements in data arbitrage, multilingual preference training, and model merging, Aya Expanse sets a new state-of-the-art in multilingual performance. Our evaluations on the Arena-Hard-Auto dataset, translated into 23 languages, demonstrate that Aya Expanse 8B and 32B outperform leading open-weight models in their respective parameter classes, including Gemma 2, Qwen 2.5, and Llama 3.1, achieving up to a 76.6% win-rate. Notably, Aya Expanse 32B outperforms Llama 3.1 70B, a model with twice as many parameters, achieving a 54.0% win-rate. In this short technical report, we present extended evaluation results for the Aya Expanse model family and release their open-weights, together with a new multilingual evaluation dataset m-ArenaHard.

Cultural biases in multilingual datasets pose significant challenges for their effectiveness as global benchmarks. These biases stem not only from language but also from the cultural knowledge required to interpret questions, reducing the practical utility of translated datasets like MMLU. Furthermore, translation often introduces artifacts that can distort the meaning or clarity of questions in the target language. A common practice in multilingual evaluation is to rely on machine-translated evaluation sets, but simply translating a dataset is insufficient to address these challenges. In this work, we trace the impact of both of these issues on multilingual evaluations and ensuing model performances. Our large-scale evaluation of state-of-the-art open and proprietary models illustrates that progress on MMLU depends heavily on learning Western-centric concepts, with 28% of all questions requiring culturally sensitive knowledge. Moreover, for questions requiring geographic knowledge, an astounding 84.9% focus on either North American or European regions. Rankings of model evaluations change depending on whether they are evaluated on the full portion or the subset of questions annotated as culturally sensitive, showing the distortion to model rankings when blindly relying on translated MMLU. We release Global-MMLU, an improved MMLU with evaluation coverage across 42 languages -- with improved overall quality by engaging with compensated professional and community annotators to verify translation quality while also rigorously evaluating cultural biases present in the original dataset. This comprehensive Global-MMLU set also includes designated subsets labeled as culturally sensitive and culturally agnostic to allow for more holistic, complete evaluation.

We present a novel multi-messenger approach for probing nonstandard neutrino properties through the detection of gravitational waves (GWs) from collapsing stellar cores and associated supernova explosions. We show that neutrino flavor conversion inside the proto-neutron star (PNS), motivated by physics Beyond the Standard Model (BSM), can significantly boost PNS convection. This effect leads to large-amplitude GW emission over a wide frequency range during an otherwise relatively quiescent GW phase shortly after core bounce. Such a signal provides a promising new avenue for exploring nonstandard neutrino phenomena and other BSM physics impacting PNS convection., Comment: 10 pages, 5 figures; submitted to Phys. Rev. Letters

Open compound domain adaptation (OCDA) is a practical domain adaptation problem that consists of a source domain, target compound domain, and unseen open domain. In this problem, the absence of domain labels and pixel-level segmentation labels for both compound and open domains poses challenges to the direct application of existing domain adaptation and generalization methods. To address this issue, we propose Amplitude-based curriculum learning and a Hopfield segmentation model for Open Compound Domain Adaptation (AH-OCDA). Our method comprises two complementary components: 1) amplitude-based curriculum learning and 2) Hopfield segmentation model. Without prior knowledge of target domains within the compound domains, amplitude-based curriculum learning gradually induces the semantic segmentation model to adapt from the near-source compound domain to the far-source compound domain by ranking unlabeled compound domain images through Fast Fourier Transform (FFT). Additionally, the Hopfield segmentation model maps segmentation feature distributions from arbitrary domains to the feature distributions of the source domain. AH-OCDA achieves state-of-the-art performance on two OCDA benchmarks and extended open domains, demonstrating its adaptability to continuously changing compound domains and unseen open domains., Comment: WACV 2025

Recent text-to-image diffusion models leverage cross-attention layers, which have been effectively utilized to enhance a range of visual generative tasks. However, our understanding of cross-attention layers remains somewhat limited. In this study, we present a method for constructing Head Relevance Vectors (HRVs) that align with useful visual concepts. An HRV for a given visual concept is a vector with a length equal to the total number of cross-attention heads, where each element represents the importance of the corresponding head for the given visual concept. We develop and employ an ordered weakening analysis to demonstrate the effectiveness of HRVs as interpretable features. To demonstrate the utility of HRVs, we propose concept strengthening and concept adjusting methods and apply them to enhance three visual generative tasks. We show that misinterpretations of polysemous words in image generation can be corrected in most cases, five challenging attributes in image editing can be successfully modified, and catastrophic neglect in multi-concept generation can be mitigated. Overall, our work provides an advancement in understanding cross-attention layers and introduces new approaches for fine-controlling these layers at the head level.

In this paper, we study two Higgs doublet models with gauged U(1)_H symmetry, motivated by the excesses around 96 GeV reported by the CMS collaboration in the searches for light resonances decaying to two photons and two \tau's. In this model, one Higgs doublet field is charged under the U(1)_H symmetry to avoid tree-level flavor changing neutral currents. The extra gauge symmetry requires extra chiral fermions, to satisfy the anomaly-free conditions. We analyze the signals of the light resonances, taking into account the contribution of the extra fermions, and discuss the consistency with the experimental results in this model., Comment: 19 pages, 5 figures

It is yet to be understood how large-scale environments influence star formation activity in galaxy clusters. One recently proposed mechanism is that galaxy clusters can remain star-forming when fed by infalling groups and star-forming galaxies from large-scale structures surrounding them (the \textit{``web-feeding model"}). Using the COSMOS2020 catalog that has half a million galaxies with high accuracy ($\sigma_{\Delta z /1+z} \sim 0.01$) photometric redshifts, we study the relationship between star formation activities in galaxy clusters and their surrounding environment to test the web-feeding model. We first identify $68$ cluster candidates at $0.3 \leq z \leq 1.4$ with halo masses at $10^{13.0} - 10^{14.5}$ \SI{}{M_{\odot}}, and the surrounding large-scale structures (LSSs) with the friends-of-friends algorithm. We find that clusters with low fractions of quiescent galaxies tend to be connected with extended LSSs as expected in the web-feeding model. We also investigated the time evolution of the web-feeding trend using the IllustrisTNG cosmological simulation. Even though no clear correlation between the quiescent galaxy fraction of galaxy clusters and the significance of LSSs around them is found in the simulation, we verify that the quiescent galaxy fractions of infallers such as groups ($M_{200} \geq 10^{12}$ \SI{}{M_{\odot}}) and galaxies ($M_{200} < 10^{12}$ \SI{}{M_{\odot}}) is smaller than the quiescent fraction of cluster members and that infallers can lower the quiescent fraction of clusters. These results imply that cluster-to-cluster variations of quiescent galaxy fraction at $z \leq 1$ can at least partially be explained by feeding materials through cosmic webs to clusters., Comment: The Astrophysical Journal, 976:154 (17pp), 2024

Deep neural networks have demonstrated remarkable performance in various vision tasks, but their success heavily depends on the quality of the training data. Noisy labels are a critical issue in medical datasets and can significantly degrade model performance. Previous clean sample selection methods have not utilized the well pre-trained features of vision foundation models (VFMs) and assumed that training begins from scratch. In this paper, we propose CUFIT, a curriculum fine-tuning paradigm of VFMs for medical image classification under label noise. Our method is motivated by the fact that linear probing of VFMs is relatively unaffected by noisy samples, as it does not update the feature extractor of the VFM, thus robustly classifying the training samples. Subsequently, curriculum fine-tuning of two adapters is conducted, starting with clean sample selection from the linear probing phase. Our experimental results demonstrate that CUFIT outperforms previous methods across various medical image benchmarks. Specifically, our method surpasses previous baselines by 5.0%, 2.1%, 4.6%, and 5.8% at a 40% noise rate on the HAM10000, APTOS-2019, BloodMnist, and OrgancMnist datasets, respectively. Furthermore, we provide extensive analyses to demonstrate the impact of our method on noisy label detection. For instance, our method shows higher label precision and recall compared to previous approaches. Our work highlights the potential of leveraging VFMs in medical image classification under challenging conditions of noisy labels., Comment: Accepted at NeurIPS 2024

The rapid advancement of generative models has introduced serious risks, including deepfake techniques for facial synthesis and editing. Traditional approaches rely on training classifiers and enhancing generalizability through various feature extraction techniques. Meanwhile, training-free detection methods address issues like limited data and overfitting by directly leveraging statistical properties from vision foundation models to distinguish between real and fake images. The current leading training-free approach, RIGID, utilizes DINOv2 sensitivity to perturbations in image space for detecting fake images, with fake image embeddings exhibiting greater sensitivity than those of real images. This observation prompts us to investigate how detection performance varies across model backbones, perturbation types, and datasets. Our experiments reveal that detection performance is closely linked to model robustness, with self-supervised (SSL) models providing more reliable representations. While Gaussian noise effectively detects general objects, it performs worse on facial images, whereas Gaussian blur is more effective due to potential frequency artifacts. To further improve detection, we introduce Contrastive Blur, which enhances performance on facial images, and MINDER (MINimum distance DetEctoR), which addresses noise type bias, balancing performance across domains. Beyond performance gains, our work offers valuable insights for both the generative and detection communities, contributing to a deeper understanding of model robustness property utilized for deepfake detection.

Protecting against cyber-threats is vital for every organization and can be done by investing in cybersecurity controls and purchasing cyber insurance. However, these are interlinked since insurance premiums could be reduced by investing more in cybersecurity controls. The expected utility theory and the prospect theory are two alternative theories explaining decision-making under risk and uncertainty, which can inform strategies for optimizing resource allocation. While the former is considered a rational approach, research has shown that most people make decisions consistent with the latter, including on insurance uptakes. We compare and contrast these two approaches to provide important insights into how the two approaches could lead to different optimal allocations resulting in differing risk exposure as well as financial costs. We introduce the concept of a risk curve and show that identifying the nature of the risk curve is a key step in deriving the optimal resource allocation.

We examine the impact of various Initial Mass Function (IMF) sampling methods on the star formation and metal enrichment histories of Ultra-Faint Dwarf (UFD) galaxy analogs. These analogs are characterized by $M_{\rm vir}\sim10^8 M_\odot$ and $M_{\ast}\lesssim10^5 M_\odot$ at $z=0$, utilizing high-resolution cosmological hydrodynamic zoom-in simulations with a gas particle mass resolution of $\sim63 M_\odot$. Specifically, we evaluate three IMF sampling techniques: the burst model, stochastic IMF sampling, and individual IMF sampling. Our results demonstrate that the choice of IMF sampling method critically affects stellar feedback dynamics, particularly supernova (SN) feedback, thus impacting the star formation and metallicity characteristics of UFD analogs. We find that simulations with stochastic IMF sampling yield UFD analogs with 40\% to 70\% higher stellar masses than those using the burst model, due to a less immediate suppression of star formation by SNe. The individual IMF method results in even greater stellar masses, 8\% to 58\% more than stochastic runs, as stars form individually and continuously. Star formation is most continuous with individual sampling, followed by stochastic, and least with the burst model, which shows the longest quenching periods. Furthermore, the individual sampling approach achieves higher metallicity stars, aligning well with observed values, unlike the lower metallicities (about 1 dex less) found in the burst and stochastic methods. This difference is attributed to the continuous star formation in individual sampling, where gas metallicity shaped by previous SN events is immediately reflected in stellar metallicity. These findings emphasize the essential role of choosing appropriate IMF sampling methods for accurately modeling the star formation and chemical evolution of UFD galaxies., Comment: 17 pages, 15 figures, Submitted to MNRAS

Large language models (LLMs) often reflect real-world biases, leading to efforts to mitigate these effects and make the models unbiased. Achieving this goal requires defining clear criteria for an unbiased state, with any deviation from these criteria considered biased. Some studies define an unbiased state as equal treatment across diverse demographic groups, aiming for balanced outputs from LLMs. However, differing perspectives on equality and the importance of pluralism make it challenging to establish a universal standard. Alternatively, other approaches propose using fact-based criteria for more consistent and objective evaluations, though these methods have not yet been fully applied to LLM bias assessments. Thus, there is a need for a metric with objective criteria that offers a distinct perspective from equality-based approaches. Motivated by this need, we introduce a novel metric to assess bias using fact-based criteria and real-world statistics. In this paper, we conducted a human survey demonstrating that humans tend to perceive LLM outputs more positively when they align closely with real-world demographic distributions. Evaluating various LLMs with our proposed metric reveals that model bias varies depending on the criteria used, highlighting the need for multi-perspective assessment., Comment: Accepted in NeurIPS 2024 Workshop on Socially Responsible Language Modelling Research (SoLaR)

Leveraging the overfitting property of deep neural networks (DNNs) is trending in video delivery systems to enhance quality within bandwidth limits. Existing approaches transmit overfitted super-resolution (SR) model streams for low-resolution (LR) bitstreams, which are used to reconstruct high-resolution (HR) videos at the decoder. Although these approaches show promising results, the huge computational costs of training a large number of video frames limit their practical applications. To overcome this challenge, we propose an efficient patch sampling method named EPS for video SR network overfitting, which identifies the most valuable training patches from video frames. To this end, we first present two low-complexity Discrete Cosine Transform (DCT)-based spatial-temporal features to measure the complexity score of each patch directly. By analyzing the histogram distribution of these features, we then categorize all possible patches into different clusters and select training patches from the cluster with the highest spatial-temporal information. The number of sampled patches is adaptive based on the video content, addressing the trade-off between training complexity and efficiency. Our method reduces the number of patches for the training to 4% to 25%, depending on the resolution and number of clusters, while maintaining high video quality and significantly enhancing training efficiency. Compared to the state-of-the-art patch sampling method, EMT, our approach achieves an 83% decrease in overall run time.

Neural networks struggle with image classification when biases are learned and misleads correlations, affecting their generalization and performance. Previous methods require attribute labels (e.g. background, color) or utilizes Generative Adversarial Networks (GANs) to mitigate biases. We introduce DiffuBias, a novel pipeline for text-to-image generation that enhances classifier robustness by generating bias-conflict samples, without requiring training during the generation phase. Utilizing pretrained diffusion and image captioning models, DiffuBias generates images that challenge the biases of classifiers, using the top-$K$ losses from a biased classifier ($f_B$) to create more representative data samples. This method not only debiases effectively but also boosts classifier generalization capabilities. To the best of our knowledge, DiffuBias is the first approach leveraging a stable diffusion model to generate bias-conflict samples in debiasing tasks. Our comprehensive experimental evaluations demonstrate that DiffuBias achieves state-of-the-art performance on benchmark datasets. We also conduct a comparative analysis of various generative models in terms of carbon emissions and energy consumption to highlight the significance of computational efficiency., Comment: 8 pages + Appendix

Diffusion models have enabled the generation of high-quality images with a strong focus on realism and textual fidelity. Yet, large-scale text-to-image models, such as Stable Diffusion, struggle to generate images where foreground objects are placed over a chroma key background, limiting their ability to separate foreground and background elements without fine-tuning. To address this limitation, we present a novel Training-Free Chroma Key Content Generation Diffusion Model (TKG-DM), which optimizes the initial random noise to produce images with foreground objects on a specifiable color background. Our proposed method is the first to explore the manipulation of the color aspects in initial noise for controlled background generation, enabling precise separation of foreground and background without fine-tuning. Extensive experiments demonstrate that our training-free method outperforms existing methods in both qualitative and quantitative evaluations, matching or surpassing fine-tuned models. Finally, we successfully extend it to other tasks (e.g., consistency models and text-to-video), highlighting its transformative potential across various generative applications where independent control of foreground and background is crucial.

Gradient-boosting decision forests, as used by algorithms such as XGBoost or AdaBoost, offer higher accuracy and lower training times for large datasets than decision trees. Protocols for private inference over decision trees can be used to preserve the privacy of the input data as well as the privacy of the trees. However, naively extending private inference over decision trees to private inference over decision forests by replicating the protocols leads to impractical running times. In this paper, we explore extending the private decision inference protocol using homomorphic encryption by Mahdavi et al. (CCS 2023) to decision forests. We present several optimizations that identify and then remove (approximate) duplication between the trees in a forest and hence achieve significant improvements in communication and computation cost over the naive approach. To the best of our knowledge, we present the first private inference protocol for highly scalable gradient-boosting decision forests. Our optimizations extend beyond Mahdavi et al.'s protocol to various private inference protocols for gradient-boosting decision trees. Our protocol's inference time is faster than the baseline of parallel running the protocol by Mahdavi et al.~by up to 28.1x, and faster than Zama's Concrete ML XGBoost by up to 122.25x.

Two enigmatic gamma-ray features in the Galactic central region, known as Fermi Bubbles (FBs), were found from Fermi-LAT data. An energy release (e.g., by tidal disruption events in the Galactic center, GC), generates a cavity with a shock that expands into the local ambient medium of the Galactic halo. A decade or so ago, a phenomenological model of the FBs was suggested as a result of routine star disruptions by the supermassive black hole in the GC which might provide enough energy for large-scale structures, like the FBs. In 2020, analytical and numerical models of the FBs as a process of routine tidal disruption of stars near the GC were developed, which can provide enough cumulative energy to form and maintain large scale structures like the FBs. The disruption events are expected to be ten to hundred events per million years, providing the average power of energy release from the GC into the halo of 3E41 erg/s, which is needed to support the FBs. Analysis of the evolution of superbubbles in exponentially stratified disks concluded that the FB envelope would be destroyed by the Rayleigh-Taylor (RT) instabilities at late stages. The shell is composed of a swept-up gas of the bubble, whose thickness is much thinner in comparison to the size of the envelope. We assume that hydrodynamic turbulence is excited in the FB envelope by the RT instability. In this case, the universal energy spectrum of turbulence may be developed in the inertial range of wavenumbers of fluctuations (the Kolmogorov-Obukhov spectrum). From our model we suppose the power of the FBs is transformed partly into the energy of hydrodynamic turbulence in the envelope. If so, hydrodynamic turbulence may generate MHD-fluctuations, which accelerate cosmic rays there and generate gamma-ray and radio emission from the FBs. We hope that this model may interpret the observed nonthermal emission from the bubbles., Comment: 21 pages, 13 figures

Reproducing the physical characteristics of ultra-faint dwarf galaxies (UFDs) in cosmological simulations is challenging, particularly with respect to stellar metallicity and galaxy size. To investigate these difficulties in detail, we conduct high-resolution simulations ($M_{\rm gas} \sim 60 \, M_{\odot}$, $M_{\rm DM} \sim 370 \, M_{\odot}$ ) on six UFD analogs ($M_{\rm vir} \sim 10^8 - 10^9 \, M_{\odot}$, $M_{\rm \star} \sim 10^3 - 2.1 \times 10^4 \, M_{\odot}$). Our findings reveal that the stellar properties of UFD analogs are shaped by diverse star-forming environments from multiple progenitor halos in the early Universe. Notably, our UFD analogs exhibit a better match to the observed mass-metallicity relation (MZR), showing higher average metallicity compared to other theoretical models. The metallicity distribution functions (MDFs) of our simulated UFDs lack high-metallicity stars ($[\rm Fe/H] > -2.0$) while containing low-metallicity stars ($[\rm Fe/H] < -4.0$). Excluding these low-metallicity stars, our results align well with the MDFs of observed UFDs. However, forming stars with higher metallicity ($-2.0 \leq [\rm Fe/H]_{\rm max} \leq -1.5$) remains a challenge due to the difficulty of sustaining metal enrichment during their brief star formation period before cosmic reionization. Additionally, our simulations show extended outer structures in UFDs, resulting from dry mergers between progenitor halos. To ensure consistency, we adopt the same fitting method commonly used in observations to derive the half-light radius. We find that this method tends to produce lower values compared to direct calculations and struggles to accurately describe the extended outer structures. To address this, we employ a two-component density profile to obtain structural parameters, finding that it better describes the galaxy shape, including both inner and outer structures., Comment: 28 pages, 21 figures

This study investigates language models' generative capabilities in tool-use dialogs. We categorize the models' outputs in tool-use dialogs into four distinct types: Tool Call, Answer Completion, Slot Question, and Relevance Detection, which serve as aspects for evaluation. We introduce FunctionChat-Bench, comprising 700 evaluation items and automated assessment programs. Using this benchmark, we evaluate several language models that support function calling. Our findings indicate that while language models may exhibit high accuracy in single-turn Tool Call scenarios, this does not necessarily translate to superior generative performance in multi-turn environments. We argue that the capabilities required for function calling extend beyond generating tool call messages; they must also effectively generate conversational messages that engage the user., Comment: 8 pages

To realize braiding of vortex lines and understand the basic properties of the energy landscape for vortex motion, precise manipulation of superconducting vortices on the nanoscale is required. Here, we reveal that a localized trapping potential powerful enough to pull in the vortex line can be created with nanoscale precision on the surface of an FeSe superconductor using the tip of a scanning tunneling microscope. The mechanism of tip-induced force is traced to local modification of electronic properties and reduction of the superconducting gap, most likely due to tip-induced strain. Intriguingly, the tip-induced trapping potential is much less pronounced along the twin boundaries, dramatically reducing the vortice's degree of motion relative to the surrounding lattice. By enabling nanoscale manipulation of single vortices in Fe-based superconductors, and likely similar materials with strong strain-susceptibility of the superconducting gap, our findings provide an important step toward further development of vortex-based quantum information processing.

Federated Learning (FL) enables collaborative model training across distributed devices while preserving local data privacy, making it ideal for mobile and embedded systems. However, the decentralized nature of FL also opens vulnerabilities to model poisoning attacks, particularly backdoor attacks, where adversaries implant trigger patterns to manipulate model predictions. In this paper, we propose DeTrigger, a scalable and efficient backdoor-robust federated learning framework that leverages insights from adversarial attack methodologies. By employing gradient analysis with temperature scaling, DeTrigger detects and isolates backdoor triggers, allowing for precise model weight pruning of backdoor activations without sacrificing benign model knowledge. Extensive evaluations across four widely used datasets demonstrate that DeTrigger achieves up to 251x faster detection than traditional methods and mitigates backdoor attacks by up to 98.9%, with minimal impact on global model accuracy. Our findings establish DeTrigger as a robust and scalable solution to protect federated learning environments against sophisticated backdoor threats., Comment: 14 pages

Robot swarms offer significant potential for inspecting diverse infrastructure, ranging from bridges to space stations. However, effective inspection requires accurate robot localization, which demands substantial computational resources and limits productivity. Inspired by biological systems, we introduce a novel cooperative localization mechanism that minimizes collective computation expenditure through self-organized sacrifice. Here, a few agents bear the computational burden of localization; through local interactions, they improve the inspection productivity of the swarm. Our approach adaptively maximizes inspection productivity for unconstrained trajectories in dynamic interaction and environmental settings. We demonstrate the optimality and robustness using mean-field analytical models, multi-agent simulations, and hardware experiments with metal climbing robots inspecting a 3D cylinder., Comment: 14 pages, 10 figures, 17th International Symposium on Distributed Autonomous Robotic Systems (DARS'24)

In this study, we propose an ensemble learning framework for electroencephalogram-based overt speech classification, leveraging denoising diffusion probabilistic models with varying convolutional kernel sizes. The ensemble comprises three models with kernel sizes of 51, 101, and 201, effectively capturing multi-scale temporal features inherent in signals. This approach improves the robustness and accuracy of speech decoding by accommodating the rich temporal complexity of neural signals. The ensemble models work in conjunction with conditional autoencoders that refine the reconstructed signals and maximize the useful information for downstream classification tasks. The results indicate that the proposed ensemble-based approach significantly outperforms individual models and existing state-of-the-art techniques. These findings demonstrate the potential of ensemble methods in advancing brain signal decoding, offering new possibilities for non-verbal communication applications, particularly in brain-computer interface systems aimed at aiding individuals with speech impairments.

We investigate the formation history of intrahalo light (IHL) using the high-resolution (~1 kpc), large-scale (~Gpc) cosmological hydrodynamical simulation, Horizon Run 5 (HR5). IHL particles are identified by carefully considering both their binding energies and positions with respect to the tidal radii of individual galaxies. By analyzing more than 1,200 galaxy groups and clusters with $\geq 10^{13} M_{\odot}$ and tracing their individual IHL particles back in time, we classify the origin of each IHL particle at each epoch based on the status of the originating galaxy into three categories: brightest halo galaxy (BHG) formation/merger, satellite galaxy stripping, and pre-processing. Our study reveals that the IHL production through BHG formation/merger is the predominant production channel, contributing over 60\% of the total IHL mass across all redshifts. The second most significant IHL production channel is pre-processing, providing more than 20\% in the final HR5 snapshot. Stripping is negligible at $z>4$ but becomes gradually more important as halos mature at $z<4$. Finally, we verify that IHL production through the disruption of dwarf galaxies and in-situ formation is negligible, contributing less than ~3\% and ~0.5\% to the total IHL production, respectively., Comment: Submitted to ApJ, 14 pages, 11 figures

The collection of speech data carried out in Sociolinguistics has the potential to enhance large language models due to its quality and representativeness. In this paper, we examine the ethical considerations associated with the gathering and dissemination of such data. Additionally, we outline strategies for addressing the sensitivity of speech data, as it may facilitate the identification of informants who contributed with their speech., Comment: in Portuguese language. Paper accepted to LAAI-Ethics 2024

We construct a locally profinite set of cardinality $\aleph_{\omega}$ with infinitely many first cohomology classes of which any distinct finite product does not vanish. Building on this, we construct the first example of a nondescendable faithfully flat map between commutative rings of cardinality $\aleph_{\omega}$ within Zermelo--Fraenkel set theory., Comment: 7 pages

A new 2,400 L liquid scintillator has been produced for the COSINE-100 Upgrade, which is under construction at Yemilab for the next COSINE dark matter experiment phase. The linear-alkyl-benzene-based scintillator is designed to serve as a veto for NaI(Tl) crystal targets and a separate platform for rare event searches. We measured using a sample consisting of a custom-made 445 mL cylindrical Teflon container equipped with two 3-inch photomultiplier tubes. Analyses show activity levels of $0.091 \pm 0.042$ mBq/kg for $^{238}$U and $0.012 \pm 0.007$ mBq/kg for $^{232}$Th.

The exploration of kilometre-sized trans-Neptunian objects (TNOs) is one of the ultimate goals in the search for the origin and evolution of the Solar System. However, such exploration is challenging because these small bodies are too faint to be directly detected. One potential avenue for detecting and investigating small and faint TNOs is the monitoring of stellar occultation events. This paper reviews the concept and methodology of monitoring observations of stellar occultations by small, unidentified TNOs, focusing on our observational programme, Organized Autotelescopes for Serendipitous Event Survey (OASES). OASES aims to detect and investigate stellar occultations by unidentified TNOs using multiple amateur-sized telescopes equipped with commercial Complementary Metal-Oxide-Semiconductor (CMOS) cameras. Through the monitoring observations conducted so far, OASES has found one possible occultation event by a kilometre-sized TNO. The paper also discusses future developments of the OASES project and deliberates on the potential of movie observations in expanding the frontiers of outer Solar System research. This article is part of the themed issue "Major Advances in Planetary Sciences thanks to Stellar Occultations"., Comment: 11 pages, 7 figures, accepted for publication in the upcoming theme issue of Philosophical Transactions A

Using optical technology for current injection and electromagnetic emission simplifies the comparison between materials. Here, we inject current into monolayer graphene and bulk gallium arsenide (GaAs) using two-color quantum interference and detect the emitted electric field by electro-optic sampling. We find the amplitude of emitted terahertz (THz) radiation scales in the same way for both materials even though they differ in dimension, band gap, atomic composition, symmetry and lattice structure. In addition, we observe the same mapping of the current direction to the light characteristics. With no electrodes for injection or detection, our approach will allow electron scattering timescales to be directly measured. We envisage that it will enable exploration of new materials suitable for generating terahertz magnetic fields., Comment: 4 figures

We show a linear sized reduction from the Maximum Cut Problem (Max-Cut) with completeness $1 - \varepsilon$ and soundness $1 - \varepsilon^{1/2}$ to the $\gamma$-Approximate Closest Vector Problem under any finite $\ell_p$-norm including $p = 2$. This reduction implies two headline results: (i) We show that any sub-exponential time (classical or quantum) algorithm for the $o(\sqrt{\log n}^{\frac{1}{p}})$-Approximate Closest Vector Problem in any finite $\ell_p$-norm implies a faster than the state-of-the-art (by Arora, Barak, and Steurer [\textit{Journal of the ACM}, 2015]) sub-exponential time (classical or quantum) algorithm for Max-Cut. This fills the gap between the results by Bennett, Golovnev, and Stephens-Davidowitz [\textit{FOCS} 2017] which had an almost optimal runtime lower bound but a very small approximation factor and the results by Dinur, Kindler, Raz, and Safra [\textit{Combinatorica}, 2003] which had an almost optimal approximation factor but small runtime lower bound, albeit using a different underlying hard problem; (ii) in combination with the classical results of Aggarwal and Kumar [\textit{FOCS} 2023] and our quantization of those results, there are no fine-grained reductions from $k$-SAT to Max-Cut with one-sided error, nor are there non-adaptive fine-grained (classical or quantum) reductions with two-sided error, unless the polynomial hierarchy collapses (or unless $\mathrm{NP} \subseteq \mathrm{pr} \text{-} \mathrm{QSZK}$ in the quantum case). The second result poses a significant barrier against proving the fine-grained complexity of Max-Cut using the Strong Exponential Time Hypothesis (or the Quantum Strong Exponential Time Hypothesis)., Comment: 35 pages, 3 figures

Due to the lack of state dimension optimization methods, deep state space models (SSMs) have sacrificed model capacity, training search space, or stability to alleviate computational costs caused by high state dimensions. In this work, we provide a structured pruning method for SSMs, Layer-Adaptive STate pruning (LAST), which reduces the state dimension of each layer in minimizing model-level energy loss by extending modal truncation for a single system. LAST scores are evaluated using $\mathcal{H}_{\infty}$ norms of subsystems for each state and layer-wise energy normalization. The scores serve as global pruning criteria, enabling cross-layer comparison of states and layer-adaptive pruning. Across various sequence benchmarks, LAST optimizes previous SSMs, revealing the redundancy and compressibility of their state spaces. Notably, we demonstrate that, on average, pruning 33% of states still maintains performance with 0.52% accuracy loss in multi-input multi-output SSMs without retraining. Code is available at $\href{https://github.com/msgwak/LAST}{\text{this https URL}}$.

Matrix denoising is central to signal processing and machine learning. Its analysis when the matrix to infer has a factorised structure with a rank growing proportionally to its dimension remains a challenge, except when it is rotationally invariant. In this case the information theoretic limits and a Bayes-optimal denoising algorithm, called rotational invariant estimator [1,2], are known. Beyond this setting few results can be found. The reason is that the model is not a usual spin system because of the growing rank dimension, nor a matrix model due to the lack of rotation symmetry, but rather a hybrid between the two. In this paper we make progress towards the understanding of Bayesian matrix denoising when the hidden signal is a factored matrix $XX^\intercal$ that is not rotationally invariant. Monte Carlo simulations suggest the existence of a denoising-factorisation transition separating a phase where denoising using the rotational invariant estimator remains Bayes-optimal due to universality properties of the same nature as in random matrix theory, from one where universality breaks down and better denoising is possible by exploiting the signal's prior and factorised structure, though algorithmically hard. We also argue that it is only beyond the transition that factorisation, i.e., estimating $X$ itself, becomes possible up to sign and permutation ambiguities. On the theoretical side, we combine mean-field techniques in an interpretable multiscale fashion in order to access the minimum mean-square error and mutual information. Interestingly, our alternative method yields equations which can be reproduced using the replica approach of [3]. Using numerical insights, we then delimit the portion of the phase diagram where this mean-field theory is reliable, and correct it using universality when it is not. Our ansatz matches well the numerics when accounting for finite size effects.

This study examines the applicability of the Cloze test, a widely used tool for assessing text comprehension proficiency, while highlighting its challenges in large-scale implementation. To address these limitations, an automated correction approach was proposed, utilizing Natural Language Processing (NLP) techniques, particularly word embeddings (WE) models, to assess semantic similarity between expected and provided answers. Using data from Cloze tests administered to students in Brazil, WE models for Brazilian Portuguese (PT-BR) were employed to measure the semantic similarity of the responses. The results were validated through an experimental setup involving twelve judges who classified the students' answers. A comparative analysis between the WE models' scores and the judges' evaluations revealed that GloVe was the most effective model, demonstrating the highest correlation with the judges' assessments. This study underscores the utility of WE models in evaluating semantic similarity and their potential to enhance large-scale Cloze test assessments. Furthermore, it contributes to educational assessment methodologies by offering a more efficient approach to evaluating reading proficiency.

Linguistic diversity is a human attribute which, with the advance of generative AIs, is coming under threat. This paper, based on the contributions of sociolinguistics, examines the consequences of the variety selection bias imposed by technological applications and the vicious circle of preserving a variety that becomes dominant and standardized because it has linguistic documentation to feed the large language models for machine learning., Comment: in Portuguese language. paper aceepted to LAAI-Ethics 2024

To automatically detect Anterior Mediastinum Lesions (AMLs) in the Anterior Mediastinum (AM), the primary requirement will be an automatic segmentation model specifically designed for the AM. The prevalence of AML is extremely low, making it challenging to conduct screening research similar to lung cancer screening. Retrospectively reviewing chest CT scans over a specific period to investigate the prevalence of AML requires substantial time. Therefore, developing an Artificial Intelligence (AI) model to find location of AM helps radiologist to enhance their ability to manage workloads and improve diagnostic accuracy for AMLs. In this paper, we introduce a U-shaped structure network to segment AM. Two attention mechanisms were used for maintaining long-range dependencies and localization. In order to have the potential of Multi-Head Self-Attention (MHSA) and a lightweight network, we designed a parallel MHSA named Wide-MHSA (W-MHSA). Maintaining long-range dependencies is crucial for segmentation when we upsample feature maps. Therefore, we designed a Dilated Depth-Wise Parallel Path connection (DDWPP) for this purpose. In order to design a lightweight architecture, we introduced an expanding convolution block and combine it with the proposed W-MHSA for feature extraction in the encoder part of the proposed U-shaped network. The proposed network was trained on 2775 AM cases, which obtained an average Dice Similarity Coefficient (DSC) of 87.83%, mean Intersection over Union (IoU) of 79.16%, and Sensitivity of 89.60%. Our proposed architecture exhibited superior segmentation performance compared to the most advanced segmentation networks, such as Trans Unet, Attention Unet, Res Unet, and Res Unet++.

Large Language Models (LLMs) have revolutionized various domains but remain vulnerable to prompt injection attacks, where malicious inputs manipulate the model into ignoring original instructions and executing designated action. In this paper, we investigate the underlying mechanisms of these attacks by analyzing the attention patterns within LLMs. We introduce the concept of the distraction effect, where specific attention heads, termed important heads, shift focus from the original instruction to the injected instruction. Building on this discovery, we propose Attention Tracker, a training-free detection method that tracks attention patterns on instruction to detect prompt injection attacks without the need for additional LLM inference. Our method generalizes effectively across diverse models, datasets, and attack types, showing an AUROC improvement of up to 10.0% over existing methods, and performs well even on small LLMs. We demonstrate the robustness of our approach through extensive evaluations and provide insights into safeguarding LLM-integrated systems from prompt injection vulnerabilities., Comment: Project page: https://huggingface.co/spaces/TrustSafeAI/Attention-Tracker

Large Language Models (LLMs) have demonstrated remarkable generalization and instruction-following capabilities with instruction tuning. The advancements in LLMs and instruction tuning have led to the development of Large Vision-Language Models (LVLMs). However, the competency of the LLMs and instruction tuning have been less explored in the molecular domain. Thus, we propose LLaMo: Large Language Model-based Molecular graph assistant, which is an end-to-end trained large molecular graph-language model. To bridge the discrepancy between the language and graph modalities, we present the multi-level graph projector that transforms graph representations into graph tokens by abstracting the output representations of each GNN layer and motif representations with the cross-attention mechanism. We also introduce machine-generated molecular graph instruction data to instruction-tune the large molecular graph-language model for general-purpose molecule and language understanding. Our extensive experiments demonstrate that LLaMo shows the best performance on diverse tasks, such as molecular description generation, property prediction, and IUPAC name prediction. The code of LLaMo is available at https://github.com/mlvlab/LLaMo., Comment: NeurIPS 2024

The schooling behavior of fish is hypothesized to confer many survival benefits, including foraging success, safety from predators, and energy savings through hydrodynamic interactions when swimming in formation. Underwater robot collectives may be able to achieve similar benefits in future applications, e.g. using formation control to achieve efficient spatial sampling for environmental monitoring. Although many theoretical algorithms exist for multi-robot formation control, they have not been tested in the underwater domain due to the fundamental challenges in underwater communication. Here we introduce a leader-follower strategy for underwater formation control that allows us to realize complex 3D formations, using purely vision-based perception and a reactive control algorithm that is low computation. We use a physical platform, BlueSwarm, to demonstrate for the first time an experimental realization of inline, side-by-side, and staggered swimming 3D formations. More complex formations are studied in a physics-based simulator, providing new insights into the convergence and stability of formations given underwater inertial/drag conditions. Our findings lay the groundwork for future applications of underwater robot swarms in aquatic environments with minimal communication., Comment: Accepted at DARS 2024 (The 17th International Symposium on Distributed Autonomous Robotic Systems)

Our work explores the utilization of deep learning, specifically leveraging the CodeBERT model, to enhance code security testing for Python applications by detecting SQL injection vulnerabilities. Unlike traditional security testing methods that may be slow and error-prone, our approach transforms source code into vector representations and trains a Long Short-Term Memory (LSTM) model to identify vulnerable patterns. When compared with existing static application security testing (SAST) tools, our model displays superior performance, achieving higher precision, recall, and F1-score. The study demonstrates that deep learning techniques, particularly with CodeBERT's advanced contextual understanding, can significantly improve vulnerability detection, presenting a scalable methodology applicable to various programming languages and vulnerability types., Comment: 4 pages, 1 figures; Presented at The 30st International Conference on Computational & Experimental Engineering and Sciences (ICCES2024)

Exploiting symmetry inherent in data can significantly improve the sample efficiency of a learning procedure and the generalization of learned models. When data clearly reveals underlying symmetry, leveraging this symmetry can naturally inform the design of model architectures or learning strategies. Yet, in numerous real-world scenarios, identifying the specific symmetry within a given data distribution often proves ambiguous. To tackle this, some existing works learn symmetry in a data-driven manner, parameterizing and learning expected symmetry through data. However, these methods often rely on explicit knowledge, such as pre-defined Lie groups, which are typically restricted to linear or affine transformations. In this paper, we propose a novel symmetry learning algorithm based on transformations defined with one-parameter groups, continuously parameterized transformations flowing along the directions of vector fields called infinitesimal generators. Our method is built upon minimal inductive biases, encompassing not only commonly utilized symmetries rooted in Lie groups but also extending to symmetries derived from nonlinear generators. To learn these symmetries, we introduce a notion of a validity score that examine whether the transformed data is still valid for the given task. The validity score is designed to be fully differentiable and easily computable, enabling effective searches for transformations that achieve symmetries innate to the data. We apply our method mainly in two domains: image data and partial differential equations, and demonstrate its advantages. Our codes are available at \url{https://github.com/kogyeonghoon/learning-symmetry-from-scratch.git}., Comment: Neurips 2024

For a compact Riemannian $3$-manifold $(M^{3}, g)$ with mean convex boundary which is diffeomorphic to a weakly convex compact domain in $\mathbb{R}^{3}$, we prove that if scalar curvature is nonnegative and the scaled mean curvature comparison $H^{2}g \ge H_{0}^{2} g_{Eucl}$ holds then $(M,g)$ is flat. Our result is a smooth analog of Gromov's dihedral rigidity conjecture and an effective version of extremality results on weakly convex balls in $\mathbb R^3$. More generally, we prove the comparison and rigidity theorem for several classes of manifold with corners. Our proof uses capillary minimal surfaces with prescribed contact angle together with the construction of foliation with nonnegative mean curvature and with prescribed contact angles., Comment: 20 pages, comments are welcome!

We investigate hysteresis in a generalized Kuramoto model with identical oscillators, focusing on coupling strength inhomogeneity, which results in oscillators being coupled to others with varying strength, and a simplified, more realistic coupling function. With the more realistic coupling function and the coupling strength inhomogeneity, each oscillator acquires an effective intrinsic frequency proportional to its individual coupling strength. This is analogous to the positive coupling strength-frequency correlation introduced explicitly or implicitly in some previous models with nonidentical oscillators that show explosive synchronization and hysteresis. Through numerical simulations and analysis using truncated Gaussian, uniform, and truncated power-law coupling strength distributions, we observe that the system can exhibit abrupt phase transitions and hysteresis. The distribution of coupling strengths significantly affects the hysteresis regions within the parameter space of the coupling function. Additionally, numerical simulations of models with weighted networks including a brain network confirm the existence of hysteresis due to the realistic coupling function and coupling strength inhomogeneity, suggesting the broad applicability of our findings to complex real-world systems., Comment: 19 pages, 8 figures

The proliferation of machine learning models in diverse clinical applications has led to a growing need for high-fidelity, medical image training data. Such data is often scarce due to cost constraints and privacy concerns. Alleviating this burden, medical image synthesis via generative adversarial networks (GANs) emerged as a powerful method for synthetically generating photo-realistic images based on existing sets of real medical images. However, the exact image set size required to efficiently train such a GAN is unclear. In this work, we experimentally establish benchmarks that measure the relationship between a sample dataset size and the fidelity of the generated images, given the dataset's distribution of image complexities. We analyze statistical metrics based on delentropy, an image complexity measure rooted in Shannon's entropy in information theory. For our pipeline, we conduct experiments with two state-of-the-art GANs, StyleGAN 3 and SPADE-GAN, trained on multiple medical imaging datasets with variable sample sizes. Across both GANs, general performance improved with increasing training set size but suffered with increasing complexity., Comment: Accepted to ACCV, Workshop on Generative AI for Synthetic Medical Data

The most massive early-type galaxies (ETGs) are known to form through numerous galaxy mergers. Thus, it is intriguing to study whether their formation in low-density environments, where nearby companions are almost absent, is associated with mergers, which are directly traced by tidal features. Using the 436 most massive ETGs with $M_\mathrm{star}>10^{11.2}\,M_{\odot}$ at $z<0.04$, we determine the variation in the fraction of massive ETGs with tidal features ($f_T$) across different environments and verify whether the most massive ETGs commonly have tidal features in very low density environments. Our main discovery is that the most massive ETGs exhibit tidal features more frequently in lower-density environments. In the highest-density environments, like galaxy clusters, $f_T$ is $0.21\pm0.06$, while in the lowest-density environments it triples to $0.62\pm0.06$. This trend is stronger for more extremely massive ETGs, with $f_T$ reaching $0.92\pm0.08$ in the lowest-density environments. One explanation for our finding is that the most massive ETGs in lower-density environments have genuinely experienced recent mergers more frequently than their counterparts in higher-density environments, suggesting that they possess extended formation histories that continue into the present. Another possibility is that tidal features last shorter in denser environments owing to external factors inherent in these environments. Our additional findings that massive ETGs with bluer $u-r$ colors are a more dominant driver of our main discovery and that dust lanes are more commonly observed in massive ETGs in low-density environments imply that gas-abundant mergers primarily contribute to the increased rate of recent mergers in low-density environments., Comment: 16 pages, 10 figures, published on October 18 in ApJ

Here we present several use cases for using Generative AI (Gen AI) to improve systems engineering and cognitive knowledge management related to the future of astronomy from a culmination of working meetings and presentations as part of the Gen AI Task Group for the NASA Habitable Worlds Observatory (HWO) Science and Technology Architecture Review Team (START) AI/ML Working Group. Collectively, our group mission statement is "Where is the Human-in-the-loop as Gen AI systems become more powerful and autonomous?" with an emphasis on the ethical applications of Gen AI, guided by using these systems to remove drudgery from human work while simultaneously increasing opportunities for humans to experience more collective creativity and innovation. The HWO mission stands to benefit dramatically from generative models for different data types including text, time series/spectra, and image data. These cover a wide range of applications in science and engineering for HWO, including: mission development acceleration, data analysis and interpretation, enhancing imaging capabilities, anomaly detection, predictive modeling and simulation, data augmentation for machine learning, instrument calibration and optimization, public engagement and education, and assisting in mission planning. As an example, through sensitivity analysis of simulated exoplanet population science data sets of various generative model complexity, we can reverse engineer the measurement uncertainty requirements for HWO instruments to produce data that can constrain population models and thus inform HWO design requirements. This approach to HWO design is one example of a strategy that can ensure that HWO remains AI-ready. Through presenting herein a combination of visionary ideas balanced with grounded validated use case examples, we aim to support the development of a long-term strategy to keep HWO AI-ready as it moves forward., Comment: Lack of guidelines for submitting work that came out of the HWO START TAG working groups.

Geometry and texture estimation from a single face image is an ill-posed problem since there is very little information to work with. The problem further escalates when the face is rotated at a different angle. This paper tries to tackle this problem by introducing a novel method for texture estimation from a single image by first using StyleGAN and 3D Morphable Models. The method begins by generating multi-view faces using the latent space of GAN. Then 3DDFA trained on 3DMM estimates a 3D face mesh as well as a high-resolution texture map that is consistent with the estimated face shape. The result shows that the generated mesh is of high quality with near to accurate texture representation., Comment: 7 pages, 4 figures, 2 tables, pre-print version

Off-axis digital holographic microscopy is a high-throughput, label-free imaging technology that provides three-dimensional, high-resolution information about samples, particularly useful in large-scale cellular imaging. However, the hologram reconstruction process poses a significant bottleneck for timely data analysis. To address this challenge, we propose a novel reconstruction approach that integrates deep learning with the physical principles of off-axis holography. We initialized part of the network weights based on the physical principle and then fine-tuned them via weakly supersized learning. Our off-axis hologram network (OAH-Net) retrieves phase and amplitude images with errors that fall within the measurement error range attributable to hardware, and its reconstruction speed significantly surpasses the microscope's acquisition rate. Crucially, OAH-Net demonstrates remarkable external generalization capabilities on unseen samples with distinct patterns and can be seamlessly integrated with other models for downstream tasks to achieve end-to-end real-time hologram analysis. This capability further expands off-axis holography's applications in both biological and medical studies., Comment: 11 pages, 4 figures