Search

As school systems work to recover from the COVID-19 pandemic, states continue to grapple with persistent teacher shortages, a lack of teachers of color in the workforce, and high turnover rates undermining recruitment efforts. Teachers' effectiveness and their likelihood of staying in teaching are strongly influenced by the quality of preparation they receive, and well-designed teacher residencies have been found to support both of these outcomes. Research suggests that teacher candidates who receive their preparation through teacher residencies--which combine comprehensive, financially supported preparation with a post-program service requirement--tend to be retained in their districts longer than other candidates, on average, thereby lowering rates of new teacher attrition and reducing the need to hire more new teachers. This research is based on the initial designs for residencies that guided the federal legislation on residencies and several early state adoptions. States and the federal government are seeking policy strategies to better recruit, prepare, and retain a qualified teacher workforce. To understand the growing evidence and policy landscape, this report begins with an overview of research on the teacher residency model and then describes state and federal policy trends and opportunities supporting teacher residencies. This includes a look across recent efforts in 12 states to fund and grow high-quality, research-aligned residencies: Arizona, California, Delaware, Indiana, Mississippi, Montana, New Mexico, New York, Pennsylvania, Texas, Washington, and West Virginia.

Recent constraints on neutron star mass and radius have advanced our understanding of the equation of state of cold dense matter. Some of them have been obtained by modeling the pulses of three millisecond X-ray pulsars observed by the Neutron Star Interior Composition Explorer (NICER). Here, we present a Bayesian parameter inference for a fourth pulsar, PSR J1231-1411, using the same technique with NICER and XMM-Newton data. When applying a broad mass-inclination prior from radio timing measurements and the emission region geometry model that can best explain the data, we find likely-converged results only when using a limited radius prior. If limiting the radius to be consistent with the previous observational constraints and equation of state analyses, we infer the radius to be $12.6 \pm 0.3$ km and the mass to be $1.04_{-0.03}^{+0.05}$ $M_\odot$, each reported as the posterior credible interval bounded by the $16\,\%$ and $84\,\%$ quantiles. If using an uninformative prior but limited between $10$ and $14$ km, we find otherwise similar results, but $R_{\mathrm{eq}} = 13.5_{-0.5}^{+0.3}$ km for the radius. In both cases, we find a non-antipodal hot region geometry where one emitting spot is at the equator or slightly above, surrounded by a large colder region, and where a non-circular hot region lies close to southern rotational pole. If using a wider radius prior, we only find solutions that fit the data significantly worse. We discuss the challenges in finding the better fitting solutions, possibly related to the weak interpulse feature in the pulse profile., Comment: 22 pages, 13 figures (2 of which are figure sets), 3 tables, accepted for publication in ApJ

Pulse profile modelling (PPM) is a technique for inferring mass, radius and hotspot properties of millisecond pulsars. PPM is now regularly used for analysis of rotation-powered millisecond pulsars (RMPs) with data from the Neutron Star Interior Composition ExploreR (NICER). Extending PPM to accreting millisecond pulsars (AMPs) is attractive, because they are a different source class featuring bright X-ray radiation from hotspots powered by accretion. In this paper, we present a modification of one of the PPM codes, X-PSI, so that it can be used for AMPs. In particular, we implement a model of an accretion disc and atmosphere model appropriate for the hotspots of AMPs, and improve the overall computational efficiency. We then test parameter recovery with synthetic NICER data in two scenarios with reasonable parameters for AMPs. We find in the first scenario, where the hotspot is large, that we are able to tightly and accurately constrain all parameters including mass and radius. In the second scenario, which is a high inclination system with a smaller hotspot, we find degeneracy between a subset of model parameters and a slight bias in the inferred mass and radius. This analysis of synthetic data lays the ground work for future analysis of AMPs with NICER data. Such an analysis could be complemented by future (joint) analysis of polarization data from the Imaging X-ray Polarimetry Explorer (IXPE)., Comment: 14 pages, 5 figures, 3 tables. Paper submitted to MNRAS

We report Bayesian inference of the mass, radius and hot X-ray emitting region properties - using data from the Neutron Star Interior Composition ExploreR (NICER) - for the brightest rotation-powered millisecond X-ray pulsar PSR J0437$\unicode{x2013}$4715. Our modeling is conditional on informative tight priors on mass, distance and binary inclination obtained from radio pulsar timing using the Parkes Pulsar Timing Array (PPTA) (Reardon et al. 2024), and we use NICER background models to constrain the non-source background, cross-checking with data from XMM-Newton. We assume two distinct hot emitting regions, and various parameterized hot region geometries that are defined in terms of overlapping circles; while simplified, these capture many of the possibilities suggested by detailed modeling of return current heating. For the preferred model identified by our analysis we infer a mass of $M = 1.418 \pm 0.037$ M$_\odot$ (largely informed by the PPTA mass prior) and an equatorial radius of $R = 11.36^{+0.95}_{-0.63}$ km, each reported as the posterior credible interval bounded by the 16% and 84% quantiles. This radius favors softer dense matter equations of state and is highly consistent with constraints derived from gravitational wave measurements of neutron star binary mergers. The hot regions are inferred to be non-antipodal, and hence inconsistent with a pure centered dipole magnetic field., Comment: 27 pages, 16 figures, + appendices. Accepted in ApJL. Data files and reproduction package available in Zenodo

We report an updated analysis of the radius, mass, and heated surface regions of the massive pulsar PSR J0740+6620 using Neutron Star Interior Composition Explorer (NICER) data from 2018 September 21 to 2022 April 21, a substantial increase in data set size compared to previous analyses. Using a tight mass prior from radio timing measurements and jointly modeling the new NICER data with XMM-Newton data, the inferred equatorial radius and gravitational mass are $12.49_{-0.88}^{+1.28}$ km and $2.073_{-0.069}^{+0.069}$ $M_\odot$ respectively, each reported as the posterior credible interval bounded by the $16\,\%$ and $84\,\%$ quantiles, with an estimated systematic error $\lesssim 0.1$ km. This result was obtained using the best computationally feasible sampler settings providing a strong radius lower limit but a slightly more uncertain radius upper limit. The inferred radius interval is also close to the $R=12.76_{-1.02}^{+1.49}$ km obtained by Dittmann et al., when they require the radius to be less than $16$ km as we do. The results continue to disfavor very soft equations of state for dense matter, with $R<11.15$ km for this high-mass pulsar excluded at the $95\,\%$ probability. The results do not depend significantly on the assumed cross-calibration uncertainty between NICER and XMM-Newton. Using simulated data that resemble the actual observations, we also show that our pipeline is capable of recovering parameters for the inferred models reported in this paper., Comment: 17 pages, 9 figures (2 of which are figure sets), 2 tables, published in ApJ

In real-life scenarios, humans seek out objects in the 3D world to fulfill their daily needs or intentions. This inspires us to introduce 3D intention grounding, a new task in 3D object detection employing RGB-D, based on human intention, such as "I want something to support my back". Closely related, 3D visual grounding focuses on understanding human reference. To achieve detection based on human intention, it relies on humans to observe the scene, reason out the target that aligns with their intention ("pillow" in this case), and finally provide a reference to the AI system, such as "A pillow on the couch". Instead, 3D intention grounding challenges AI agents to automatically observe, reason and detect the desired target solely based on human intention. To tackle this challenge, we introduce the new Intent3D dataset, consisting of 44,990 intention texts associated with 209 fine-grained classes from 1,042 scenes of the ScanNet dataset. We also establish several baselines based on different language-based 3D object detection models on our benchmark. Finally, we propose IntentNet, our unique approach, designed to tackle this intention-based detection problem. It focuses on three key aspects: intention understanding, reasoning to identify object candidates, and cascaded adaptive learning that leverages the intrinsic priority logic of different losses for multiple objective optimization.

Pulse profile modelling (PPM) is a comprehensive relativistic ray-tracing technique employed to determine the properties of neutron stars. In this study, we apply this technique to the Type I X-ray burster and accretion-powered millisecond pulsar XTE J1814-338, extracting its fundamental properties using PPM of its thermonuclear burst oscillations. Using data from its 2003 outburst, and a single uniform temperature hot spot model, we infer XTE J1814-338 to be located at a distance of $7.2^{+0.3}_{-0.4}$ kpc, with a mass of $1.21^{+0.05}_{-0.05}$ M$_\odot$ and an equatorial radius of $7.0^{+0.4}_{-0.4}$ km. Our results also offer insight into the time evolution of the hot spot but point to some potential shortcomings of the single uniform temperature hot spot model. We explore the implications of this result, including what we can learn about thermonuclear burst oscillation mechanisms and the importance of modelling the accretion contribution to the emission during the burst., Comment: Accepted for publication in MNRAS. The Zenodo link is public

FastCAR is a novel task consolidation approach in Multi-Task Learning (MTL) for a classification and a regression task, despite task heterogeneity with only subtle correlation. It addresses object classification and continuous property variable regression, a crucial use case in science and engineering. FastCAR involves a labeling transformation approach that can be used with a single-task regression network architecture. FastCAR outperforms traditional MTL model families, parametrized in the landscape of architecture and loss weighting schemes, when learning of both tasks are collectively considered (classification accuracy of 99.54%, regression mean absolute percentage error of 2.3%). The experiments performed used an Advanced Steel Property dataset contributed by us. The dataset comprises 4536 images of 224x224 pixels, annotated with object classes and hardness properties that take continuous values. With the labeling transformation and single-task regression network architecture, FastCAR achieves reduced latency and time efficiency.

Learning from preference-based feedback has recently gained traction as a promising approach to align language models with human interests. While these aligned generative models have demonstrated impressive capabilities across various tasks, their dependence on high-quality human preference data poses a bottleneck in practical applications. Specifically, noisy (incorrect and ambiguous) preference pairs in the dataset might restrict the language models from capturing human intent accurately. While practitioners have recently proposed heuristics to mitigate the effect of noisy preferences, a complete theoretical understanding of their workings remain elusive. In this work, we aim to bridge this gap by by introducing a general framework for policy optimization in the presence of random preference flips. We focus on the direct preference optimization (DPO) algorithm in particular since it assumes that preferences adhere to the Bradley-Terry-Luce (BTL) model, raising concerns about the impact of noisy data on the learned policy. We design a novel loss function, which de-bias the effect of noise on average, making a policy trained by minimizing that loss robust to the noise. Under log-linear parameterization of the policy class and assuming good feature coverage of the SFT policy, we prove that the sub-optimality gap of the proposed robust DPO (rDPO) policy compared to the optimal policy is of the order $O(\frac{1}{1-2\epsilon}\sqrt{\frac{d}{n}})$, where $\epsilon < 1/2$ is flip rate of labels, $d$ is policy parameter dimension and $n$ is size of dataset. Our experiments on IMDb sentiment generation and Anthropic's helpful-harmless dataset show that rDPO is robust to noise in preference labels compared to vanilla DPO and other heuristics proposed by practitioners.

Teacher shortages have reentered the national consciousness in a major way, as quarantines and the intense stresses created by the COVID-19 pandemic drained teaching staffs, causing some schools to close temporarily for lack of staff. According to the U.S. Department of Education, all 50 states reported shortages in more than one area for the 2022-23 school year. To handle the shortages, schools have increased class sizes, canceled course offerings, added duties to the responsibilities of existing teachers, and hired underqualified individuals to fill the positions--all of which undermine students' learning. At a time when the nation is necessarily focused on learning recovery for students impacted by the pandemic, resolving shortages should be a national education priority. The moment may be ripe for the set of transformations needed to develop a much stronger education profession in this country. The United States needs a nationwide Marshall Plan for teaching, similar to that enacted after World War II to rebuild Europe, but for rebuilding the teaching profession. A Marshall Plan for teaching should focus the powers of the federal government on supporting recruitment, preparation, support, and retention in teaching in seven key areas: (1) Increase educators' net compensation through tax credits, housing subsidies, and salary; (2) Strengthen recruitment by making teacher preparation debt-free; (3) Support improved preparation by expanding high-retention pathways into teaching; (4) Provide high-quality mentoring for all beginning teachers; (5) Increase investments that enable educators to expand and share expertise; (6) Incentivize the redesign of schools to support teaching and learning; and (7) Rethink school accountability. [This research was supported by Skyline Foundation.]

Over the past decade, California has revised its standards for teacher preparation and credentialing and invested in high-retention pathways for entering teaching. As part of its new accreditation system, the California Commission on Teacher Credentialing (CTC) administers surveys to program completers who apply for their preliminary teaching credentials. This analysis examines survey responses of almost 60,000 completers from 2016-17 to 2020-21. California has a growing and increasingly diverse pool of teacher preparation graduates, and more than 90% rated their programs positively. Clinical support and access to subject-area preparation are strong predictors of overall feelings of preparedness. Graduates of new preservice residencies and student teaching programs report feeling better prepared than those entering as interns or on emergency-style permits. However, access to higher-rated programs offering more clinical support varies, with half of Black and Native American candidates, as well as most special education candidates, entering without access to student teaching. [Report originally published March 2, 2023, last revised May 12, 2023. For the full report, see ED630345.]

Recently, sequential recommendation has been adapted to the LLM paradigm to enjoy the power of LLMs. LLM-based methods usually formulate recommendation information into natural language and the model is trained to predict the next item in an auto-regressive manner. Despite their notable success, the substantial computational overhead of inference poses a significant obstacle to their real-world applicability. In this work, we endeavor to streamline existing LLM-based recommendation models and propose a simple yet highly effective model Lite-LLM4Rec. The primary goal of Lite-LLM4Rec is to achieve efficient inference for the sequential recommendation task. Lite-LLM4Rec circumvents the beam search decoding by using a straight item projection head for ranking scores generation. This design stems from our empirical observation that beam search decoding is ultimately unnecessary for sequential recommendations. Additionally, Lite-LLM4Rec introduces a hierarchical LLM structure tailored to efficiently handle the extensive contextual information associated with items, thereby reducing computational overhead while enjoying the capabilities of LLMs. Experiments on three publicly available datasets corroborate the effectiveness of Lite-LLM4Rec in both performance and inference efficiency (notably 46.8% performance improvement and 97.28% efficiency improvement on ML-1m) over existing LLM-based methods. Our implementations will be open sourced., Comment: 8 pages, 5 figures, conference

An optomechanical microcavity can considerably enhance the interaction between light and mechanical motion by confining light to a sub-wavelength volume. However, this comes at the cost of an increased optical loss rate. Therefore, microcavity-based optomechanical systems are placed in the unresolved-sideband regime, preventing sideband-based ground-state cooling. A pathway to reduce optical loss in such systems is to engineer the cavity mirrors, i.e., the optical modes that interact with the mechanical resonator. In our work, we analyze such an optomechanical system, whereby one of the mirrors is strongly frequency-dependent, i.e., a suspended Fano mirror. This optomechanical system consists of two optical modes that couple to the motion of the suspended Fano mirror. We formulate a quantum-coupled-mode description that includes both the standard dispersive optomechanical coupling as well as dissipative coupling. We solve the Langevin equations of the system dynamics in the linear regime showing that ground-state cooling from room temperature can be achieved even if the cavity is per se not in the resolved-sideband regime, but achieves effective sideband resolution through strong optical mode coupling. Importantly, we find that the cavity output spectrum needs to be properly analyzed with respect to the effective laser detuning to infer the phonon occupation of the mechanical resonator. Our work also predicts how to reach the regime of nonlinear quantum optomechanics in a Fano-based microcavity by engineering the properties of the Fano mirror., Comment: 20 pages, 13 figures. Minor revision: clarify some assumptions, check energy equipartition (new fig. 13). Code to plot the article's figures available at https://gitlab.com/juliette.monsel/fano-mirror-cavity-optomechanics

We demonstrate dynamic control of the lattice by THz light by exploiting the coupling of phonon modes. The low-energy sliding phonon modes in the spin-ladder system, Sr$_{14}$Cu$_{24}$O$_{41}$, are excited using THz radiation with high electric fields. Due to the nonlinearities induced by the THz electric fields, the low-energy phonon mode couples to the higher energy, silent optical phonon modes at $\approx$ 1.17 THz. This indirect excitation of the silent phonon mode is reflected as an enhancement of the THz transmission near 1.17 THz. Our results demonstrate that it is possible to indirectly control inaccessible phonon modes with THz frequency electromagnetic radiation, which provides opportunities to alter such material systems' electronic and magnetic properties dynamically using THz radiation.

Given a query and a document corpus, the information retrieval (IR) task is to output a ranked list of relevant documents. Combining large language models (LLMs) with embedding-based retrieval models, recent work shows promising results on the zero-shot retrieval problem, i.e., no access to labeled data from the target domain. Two such popular paradigms are generation-augmented retrieval or GAR (generate additional context for the query and then retrieve), and retrieval-augmented generation or RAG (retrieve relevant documents as context and then generate answers). The success of these paradigms hinges on (i) high-recall retrieval models, which are difficult to obtain in the zero-shot setting, and (ii) high-precision (re-)ranking models which typically need a good initialization. In this work, we propose a novel GAR-meets-RAG recurrence formulation that overcomes the challenges of existing paradigms. Our method iteratively improves retrieval (via GAR) and rewrite (via RAG) stages in the zero-shot setting. A key design principle is that the rewrite-retrieval stages improve the recall of the system and a final re-ranking stage improves the precision. We conduct extensive experiments on zero-shot passage retrieval benchmarks, BEIR and TREC-DL. Our method establishes a new state-of-the-art in the BEIR benchmark, outperforming previous best results in Recall@100 and nDCG@10 metrics on 6 out of 8 datasets, with up to 17% relative gains over the previous best., Comment: preprint

Action recognition from an egocentric viewpoint is a crucial perception task in robotics and enables a wide range of human-robot interactions. While most computer vision approaches prioritize the RGB camera, the Depth modality - which can further amplify the subtleties of actions from an egocentric perspective - remains underexplored. Our work focuses on recognizing actions from egocentric RGB and Depth modalities in an industry-like environment. To study this problem, we consider the recent MECCANO dataset, which provides a wide range of assembling actions. Our framework is based on the 3D Video SWIN Transformer to encode both RGB and Depth modalities effectively. To address the inherent skewness in real-world multimodal action occurrences, we propose a training strategy using an exponentially decaying variant of the focal loss modulating factor. Additionally, to leverage the information in both RGB and Depth modalities, we opt for late fusion to combine the predictions from each modality. We thoroughly evaluate our method on the action recognition task of the MECCANO dataset, and it significantly outperforms the prior work. Notably, our method also secured first place at the multimodal action recognition challenge at ICIAP 2023.

Pulse profile modelling is a relativistic ray-tracing technique that can be used to infer masses, radii and geometric parameters of neutron stars. In a previous study, we looked at the performance of this technique when applied to thermonuclear burst oscillations from accreting neutron stars. That study showed that ignoring the variability associated with burst oscillation sources resulted in significant biases in the inferred mass and radius, particularly for the high count rates that are nominally required to obtain meaningful constraints. In this follow-on study, we show that the bias can be mitigated by slicing the bursts into shorter segments where variability can be neglected, and jointly fitting the segments. Using this approach, the systematic uncertainties on the mass and radius are brought within the range of the statistical uncertainty. With about 10$^6$ source counts, this yields uncertainties of approximately 10% for both the mass and radius. However, this modelling strategy requires substantial computational resources. We also confirm that the posterior distributions of the mass and radius obtained from multiple bursts of the same source can be merged to produce outcomes comparable to that of a single burst with an equivalent total number of counts., Comment: Accepted for publication in MNRAS. The Zenodo link is now public

In 2019 the NICER collaboration published the first mass and radius inferred for PSR J0030+0451, thanks to NICER observations, and consequent constraints on the equation of state characterising dense matter. Two independent analyses found a mass of $\sim 1.3-1.4\,\mathrm{M_\odot}$ and a radius of $\sim 13\,$km. They also both found that the hot spots were all located on the same hemisphere, opposite to the observer, and that at least one of them had a significantly elongated shape. Here we reanalyse, in greater detail, the same NICER data set, incorporating the effects of an updated NICER response matrix and using an upgraded analysis framework. We expand the adopted models and jointly analyse also XMM-Newton data, which enables us to better constrain the fraction of observed counts coming from PSR J0030+0451. Adopting the same models used in previous publications, we find consistent results, although with more stringent inference requirements. We also find a multi-modal structure in the posterior surface. This becomes crucial when XMM-Newton data is accounted for. Including the corresponding constraints disfavors the main solutions found previously, in favor of the new and more complex models. These have inferred masses and radii of $\sim [1.4 \mathrm{M_\odot}, 11.5$ km] and $\sim [1.7 \mathrm{M_\odot}, 14.5$ km], depending on the assumed model. They display configurations that do not require the two hot spots generating the observed X-rays to be on the same hemisphere, nor to show very elongated features, and point instead to the presence of temperature gradients and the need to account for them., Comment: 35 pages, 11 figures, 7 tables

We present an analysis of the effects of uncertainties in the atmosphere models on the radius, mass, and other neutron star parameter constraints for the NICER observations of rotation-powered millisecond pulsars. To date, NICER has applied the X-ray pulse profile modeling technique to two millisecond-period pulsars: PSR J0030+0451 and the high-mass pulsar PSR J0740+6620. These studies have commonly assumed a deep-heated, fully ionized hydrogen atmosphere model, although they have explored the effects of partial-ionization and helium composition in some cases. Here, we extend that exploration and also include new models with partially ionized carbon composition, externally heated hydrogen, and an empirical atmospheric beaming parameterization to explore deviations in the expected anisotropy of the emitted radiation. None of the studied atmosphere cases have any significant influence on the inferred radius of PSR J0740+6620, possibly due to its X-ray faintness, tighter external constraints, and/or viewing geometry. In the case of PSR J0030+0451, both the composition and ionization state could significantly alter the inferred radius. However, based on the evidence (prior predictive probability of the data), partially ionized hydrogen and carbon atmospheres are disfavored. The difference in the evidence for ionized hydrogen and helium atmospheres is too small to be decisive for most cases, but the inferred radius for helium models trends to larger sizes around or above 14-15 km. External heating or deviations in the beaming that are less than $5\,\%$ at emission angles smaller than 60 degrees, on the other hand, have no significant effect on the inferred radius., Comment: 26 pages, 12 figures (2 of which are figure sets), 3 tables. Published in ApJ

In the last few years, the NICER collaboration has provided mass and radius inferences, via pulse profile modeling, for two pulsars: PSR J0030+0451 and PSR J0740+6620. Given the importance of these results for constraining the equation of state of dense nuclear matter, it is crucial to validate them and test their robustness. We therefore explore the reliability of these results and their sensitivity to analysis settings and random processes, including noise, focusing on the specific case of PSR J0030+0451. We use X-PSI, one of the two main analysis pipelines currently employed by the NICER collaboration for mass and radius inferences. With synthetic data that mimic the PSR J0030+0451 NICER data set, we evaluate the recovery performances of X-PSI under conditions never tested before, including complex modeling of the thermally emitting neutron star surface. For the test cases explored, our results suggest that X-PSI is capable of recovering the true mass and radius within reasonable credible intervals. This work also reveals the main vulnerabilities of the analysis: a significant dependence on noise and the presence of multi-modal structure in the posterior surface. Noise particularly impacts our sensitivity to the analysis settings and widths of the posterior distributions. The multi-modal structure in the posterior suggests that biases could be present if the analysis is unable to exhaustively explore the parameter space. Convergence testing, to ensure an adequate coverage of the parameter space and a suitable representation of the posterior distribution, is one possible solution to these challenges., Comment: 27 pages, 13 figures

In this work, we propose an ensemble modeling approach for multimodal action recognition. We independently train individual modality models using a variant of focal loss tailored to handle the long-tailed distribution of the MECCANO [21] dataset. Based on the underlying principle of focal loss, which captures the relationship between tail (scarce) classes and their prediction difficulties, we propose an exponentially decaying variant of focal loss for our current task. It initially emphasizes learning from the hard misclassified examples and gradually adapts to the entire range of examples in the dataset. This annealing process encourages the model to strike a balance between focusing on the sparse set of hard samples, while still leveraging the information provided by the easier ones. Additionally, we opt for the late fusion strategy to combine the resultant probability distributions from RGB and Depth modalities for final action prediction. Experimental evaluations on the MECCANO dataset demonstrate the effectiveness of our approach., Comment: 22nd International Conference on Image Analysis and Processing Workshops - Multimodal Action Recognition on the MECCANO Dataset, 2023

Age prediction is an important part of medical assessments and research. It can aid in detecting diseases as well as abnormal ageing by highlighting potential discrepancies between chronological and biological age. To improve understanding of age-related changes in various body parts, we investigate the ageing of the human body on a large scale by using whole-body 3D images. We utilise the Grad-CAM method to determine the body areas most predictive of a person's age. In order to expand our analysis beyond individual subjects, we employ registration techniques to generate population-wide importance maps that show the most predictive areas in the body for a whole cohort of subjects. We show that the investigation of the full 3D volume of the whole body and the population-wide analysis can give important insights into which body parts play the most important roles in predicting a person's age. Our findings reveal three primary areas of interest: the spine, the autochthonous back muscles, and the cardiac region, which exhibits the highest importance. Finally, we investigate differences between subjects that show accelerated and decelerated ageing.

We present a detailed study of mechanically compliant, photonic-crystal-based microcavities featuring a quasi-bound state in the continuum. Such systems have recently been predicted to reduce the optical loss in Fabry-Perot-type optomechanical cavities. However, they require two identical photonic-crystal slabs facing each other, which poses a considerable challenge for experimental implementation. We investigate how such an ideal system can be simplified and still exhibit a quasi-bound state in the continuum. We find that a suspended photonic-crystal slab facing a distributed Bragg reflector realizes an optomechanical system with a quasi-bound state in the continuum. In this system, the radiative cavity loss can be eliminated to the extent that the cavity loss is dominated by dissipative loss originating from material absorption only. These proposed optomechanical cavity designs are predicted to feature optical quality factors in excess of 10^5., Comment: 11 pages, 14 figures

Supervised contrastive loss (SCL) is a competitive and often superior alternative to the cross-entropy loss for classification. While prior studies have demonstrated that both losses yield symmetric training representations under balanced data, this symmetry breaks under class imbalances. This paper presents an intriguing discovery: the introduction of a ReLU activation at the final layer effectively restores the symmetry in SCL-learned representations. We arrive at this finding analytically, by establishing that the global minimizers of an unconstrained features model with SCL loss and entry-wise non-negativity constraints form an orthogonal frame. Extensive experiments conducted across various datasets, architectures, and imbalance scenarios corroborate our finding. Importantly, our experiments reveal that the inclusion of the ReLU activation restores symmetry without compromising test accuracy. This constitutes the first geometry characterization of SCL under imbalances. Additionally, our analysis and experiments underscore the pivotal role of batch selection strategies in representation geometry. By proving necessary and sufficient conditions for mini-batch choices that ensure invariant symmetric representations, we introduce batch-binding as an efficient strategy that guarantees these conditions hold., Comment: change of title and additional experimental results