Search

Model personalization allows a set of individuals, each facing a different learning task, to train models that are more accurate for each person than those they could develop individually. The goals of personalization are captured in a variety of formal frameworks, such as multitask learning and metalearning. Combining data for model personalization poses risks for privacy because the output of an individual's model can depend on the data of other individuals. In this work we undertake a systematic study of differentially private personalized learning. Our first main contribution is to construct a taxonomy of formal frameworks for private personalized learning. This taxonomy captures different formal frameworks for learning as well as different threat models for the attacker. Our second main contribution is to prove separations between the personalized learning problems corresponding to different choices. In particular, we prove a novel separation between private multitask learning and private metalearning.

Estimating uncertainty in Large Language Models (LLMs) is important for properly evaluating LLMs, and ensuring safety for users. However, prior approaches to uncertainty estimation focus on the final answer in generated text, ignoring intermediate steps that might dramatically impact the outcome. We hypothesize that there exist key forking tokens, such that re-sampling the system at those specific tokens, but not others, leads to very different outcomes. To test this empirically, we develop a novel approach to representing uncertainty dynamics across individual tokens of text generation, and applying statistical models to test our hypothesis. Our approach is highly flexible: it can be applied to any dataset and any LLM, without fine tuning or accessing model weights. We use our method to analyze LLM responses on 7 different tasks across 4 domains, spanning a wide range of typical use cases. We find many examples of forking tokens, including surprising ones such as punctuation marks, suggesting that LLMs are often just a single token away from saying something very different.

Illusions are entertaining, but they are also a useful diagnostic tool in cognitive science, philosophy, and neuroscience. A typical illusion shows a gap between how something "really is" and how something "appears to be", and this gap helps us understand the mental processing that lead to how something appears to be. Illusions are also useful for investigating artificial systems, and much research has examined whether computational models of perceptions fall prey to the same illusions as people. Here, I invert the standard use of perceptual illusions to examine basic processing errors in current vision language models. I present these models with illusory-illusions, neighbors of common illusions that should not elicit processing errors. These include such things as perfectly reasonable ducks, crooked lines that truly are crooked, circles that seem to have different sizes because they are, in fact, of different sizes, and so on. I show that many current vision language systems mistakenly see these illusion-illusions as illusions. I suggest that such failures are part of broader failures already discussed in the literature., Comment: 9 pages, 5 figures

This work presents a comprehensive framework for the efficient implementation of finite-volume-based reacting flow solvers, specifically tailored for high speed propulsion applications. Using the exascale computing project (ECP) based AMReX framework, a compressible flow solver for handling high-speed reacting flows is developed. This work is complementary to the existing PeleC solver, emphasizing specific applications that include confined shock-containing flows, stationary and moving shocks and detonations. The framework begins with a detailed exposition of the numerical methods employed, emphasizing their application to complex geometries and their effectiveness in ensuring accurate and stable numerical simulations. Subsequently, an in-depth analysis evaluates the solver's performance across canonical and practical geometries, with particular focus on computational cost and efficiency. The solver's scalability and robustness are demonstrated through practical test cases, including flow path simulations of scramjet engines and detailed analysis of various detonation phenomena., Comment: 2024 12 01 V1

Large language models (LLMs) have demonstrated remarkable capabilities, but their adoption is limited by high computational costs during inference. While increasing parameter counts enhances accuracy, it also widens the gap between state-of-the-art capabilities and practical deployability. We present Puzzle, a framework to accelerate LLM inference on specific hardware while preserving their capabilities. Through an innovative application of neural architecture search (NAS) at an unprecedented scale, Puzzle systematically optimizes models with tens of billions of parameters under hardware constraints. Our approach utilizes blockwise local knowledge distillation (BLD) for parallel architecture exploration and employs mixed-integer programming for precise constraint optimization. We demonstrate the real-world impact of our framework through Llama-3.1-Nemotron-51B-Instruct (Nemotron-51B), a publicly available model derived from Llama-3.1-70B-Instruct. Nemotron-51B achieves a 2.17x inference throughput speedup, fitting on a single NVIDIA H100 GPU while preserving 98.4% of the original model's capabilities. Nemotron-51B currently stands as the most accurate language model capable of inference on a single GPU with large batch sizes. Remarkably, this transformation required just 45B training tokens, compared to over 15T tokens used for the 70B model it was derived from. This establishes a new paradigm where powerful models can be optimized for efficient deployment with only negligible compromise of their capabilities, demonstrating that inference performance, not parameter count alone, should guide model selection. With the release of Nemotron-51B and the presentation of the Puzzle framework, we provide practitioners immediate access to state-of-the-art language modeling capabilities at significantly reduced computational costs.

Despite remarkable progress in multi-modal AI research, there is a salient domain in which modern AI continues to lag considerably behind even human children: the reliable deployment of logical operators. Here, we examine three forms of logical operators: relations, negations, and discrete numbers. We asked human respondents (N=178 in total) to evaluate images generated by a state-of-the-art image-generating AI (DALL-E 3) prompted with these `logical probes', and find that none reliably produce human agreement scores greater than 50\%. The negation probes and numbers (beyond 3) fail most frequently. In a 4th experiment, we assess a `grounded diffusion' pipeline that leverages targeted prompt engineering and structured intermediate representations for greater compositional control, but find its performance is judged even worse than that of DALL-E 3 across prompts. To provide further clarity on potential sources of success and failure in these text-to-image systems, we supplement our 4 core experiments with multiple auxiliary analyses and schematic diagrams, directly quantifying, for example, the relationship between the N-gram frequency of relational prompts and the average match to generated images; the success rates for 3 different prompt modification strategies in the rendering of negation prompts; and the scalar variability / ratio dependence (`approximate numeracy') of prompts involving integers. We conclude by discussing the limitations inherent to `grounded' multimodal learning systems whose grounding relies heavily on vector-based semantics (e.g. DALL-E 3), or under-specified syntactical constraints (e.g. `grounded diffusion'), and propose minimal modifications (inspired by development, based in imagery) that could help to bridge the lingering compositional gap between scale and structure. All data and code is available at https://github.com/ColinConwell/T2I-Probology

Vision-Language Models (VLMs) have shown remarkable capabilities across diverse visual tasks, including image recognition, video understanding, and Visual Question Answering (VQA) when explicitly trained for these tasks. Despite these advances, we find that current VLMs lack a fundamental cognitive ability: learning to localize objects in a scene by taking into account the context. In this work, we focus on the task of few-shot personalized localization, where a model is given a small set of annotated images (in-context examples) -- each with a category label and bounding box -- and is tasked with localizing the same object type in a query image. To provoke personalized localization abilities in models, we present a data-centric solution that fine-tunes them using carefully curated data from video object tracking datasets. By leveraging sequences of frames tracking the same object across multiple shots, we simulate instruction-tuning dialogues that promote context awareness. To reinforce this, we introduce a novel regularization technique that replaces object labels with pseudo-names, ensuring the model relies on visual context rather than prior knowledge. Our method significantly enhances few-shot localization performance without sacrificing generalization, as demonstrated on several benchmarks tailored to personalized localization. This work is the first to explore and benchmark personalized few-shot localization for VLMs, laying a foundation for future research in context-driven vision-language applications. The code for our project is available at https://github.com/SivanDoveh/IPLoc

Researchers in social science and psychology have recently proposed using large language models (LLMs) as replacements for humans in behavioral research. In addition to arguments about whether LLMs accurately capture population-level patterns, this has raised questions about whether LLMs capture human-like conceptual diversity. Separately, it is debated whether post-training alignment (RLHF or RLAIF) affects models' internal diversity. Inspired by human studies, we use a new way of measuring the conceptual diversity of synthetically-generated LLM "populations" by relating the internal variability of simulated individuals to the population-level variability. We use this approach to evaluate non-aligned and aligned LLMs on two domains with rich human behavioral data. While no model reaches human-like diversity, aligned models generally display less diversity than their instruction fine-tuned counterparts. Our findings highlight potential trade-offs between increasing models' value alignment and decreasing the diversity of their conceptual representations., Comment: 17 pages, 10 figures; corrected figure version

Near-sun sky twilight observations allow for the detection of asteroid interior to the orbit of Venus (Aylos), the Earth (Atiras), and comets. We present the results of observations with the Palomar 48-inch telescope (P48)/Zwicky Transient Facility (ZTF) camera in 30 s r-band exposures taken during evening astronomical twilight from 2019 Sep 20 to 2022 March 7 and during morning astronomical twilight sky from 2019 Sep 21 to 2022 Sep 29. More than 46,000 exposures were taken in evening and morning astronomical twilight within 31 to 66 degrees from the Sun with an r-band limiting magnitude between 18.1 and 20.9. The twilight pointings show a slight seasonal dependence in limiting magnitude and ability to point closer towards the Sun, with limiting magnitude slightly improving during summer. In total, the one Aylo, (594913) 'Ayl\'o'chaxnim, and 4 Atiras, 2020 OV1, 2021 BS1, 2021 PB2, and 2021 VR3, were discovered in evening and morning twilight observations. Additional twilight survey discoveries also include 6 long-period comets: C/2020 T2, C/2020 V2, C/2021 D2, C/2021 E3, C/2022 E3, and C/2022 P3, and two short-period comets: P/2021 N1 and P/2022 P2 using deep learning comet detection pipelines. The P48/ZTF twilight survey also recovered 11 known Atiras, one Aylo, three short-period comes, two long-period comets, and one interstellar object. Lastly, the Vera Rubin Observatory will conduct a twilight survey starting in its first year of operations and will cover the sky within 45 degrees of the Sun. Twilight surveys such as those by ZTF and future surveys will provide opportunities for discovering asteroids inside the orbits of Earth and Venus., Comment: 26 pages, 13 figures, 4 tables, accepted for publication in Icarus

This study presents a novel algorithm based on machine learning (ML) for the precise segmentation and measurement of detonation cells from soot foil images, addressing the limitations of manual and primitive edge detection methods prevalent in the field. Using advances in cellular biology segmentation models, the proposed algorithm is designed to accurately extract cellular patterns without a training procedure or dataset, which is a significant challenge in detonation research. The algorithm's performance was validated using a series of test cases that mimic experimental and numerical detonation studies. The results demonstrated consistent accuracy, with errors remaining within 10%, even in complex cases. The algorithm effectively captured key cell metrics such as cell area and span, revealing trends across different soot foil samples with uniform to highly irregular cellular structures. Although the model proved robust, challenges remain in segmenting and analyzing highly complex or irregular cellular patterns. This work highlights the broad applicability and potential of the algorithm to advance the understanding of detonation wave dynamics., Comment: 23 pages, 12 figures, submitted to Comb. and Flame; v2 - added section

Planet formation is expected to be severely limited in disks of low metallicity, owing to both the small solid mass reservoir and the low opacity accelerating the disk gas dissipation. While previous studies have found a weak correlation between the occurrence rates of small planets ($\leq$4R$_\oplus$) and stellar metallicity, so far no studies have probed below the metallicity limit beyond which planet formation is predicted to be suppressed. Here, we constructed a large catalog of ~110,000 metal-poor stars observed by the TESS mission with spectroscopically-derived metallicities, and systematically probed planet formation within the metal-poor regime ([Fe/H] $\leq$ -0.5) for the first time. Extrapolating known higher-metallicity trends for small, short-period planets predicts the discovery of ~68 superEarths around these stars (~85,000 stars) after accounting for survey completeness; however, we detect none. As a result, we have placed the most stringent upper limit on super-Earth occurrence rates around metal-poor stars (-0.75 < [Fe/H] $\leq$ -0.5) to date, $\leq$ 1.67%, a statistically significant (p-value=0.000685) deviation from the prediction of metallicity trends derived with Kepler and K2. We find a clear host star metallicity cliff for super-Earths that could indicate the threshold below which planets are unable to grow beyond an Earth-mass at short orbital periods. This finding provides a crucial input to planet formation theories, and has implications for the small planet inventory of the Galaxy and the galactic epoch at which the formation of small planets started., Comment: Accepted to AJ

Multi-planet systems are a perfect laboratory for constraining planetary formation models. A few of these systems present planets that come very close to mean motion resonance, potentially leading to significant transit-timing variations (TTVs) due to their gravitational interactions. Of these systems, K2-138 represents a excellent laboratory for studying the dynamics of its six small planets (with radii ranging between $\sim1.5$ -- $3.3 R_\oplus$), as the five innermost planets are in a near 3:2 resonant chain. In this work, we aim to constrain the orbital properties of the six planets in the K2-138 system by monitoring their transits with CHaracterising ExOPlanets Satellite (CHEOPS). We also seek to use this new data to lead a TTV study on this system. We obtained twelve light curves of the system with transits of planets $d$, $e$, $f,$ and $g$. With these data, we were able to update the ephemerides of the transits for these planets and search for timing transit variations. With our measurements, we reduced the uncertainties in the orbital periods of the studied planets, typically by an order of magnitude. This allowed us to correct for large deviations, on the order of hours, in the transit times predicted by previous studies. This is key to enabling future reliable observations of the planetary transits in the system. We also highlight the presence of potential TTVs ranging from 10 minutes to as many as 60 minutes for planet $d$., Comment: Accepted in A&A

Numerical simulations of detonation-containing flows have emerged as crucial tools for designing next-generation power and propulsion devices. As these tools mature, it is important for the combustion community to properly understand and isolate grid resolution effects when simulating detonations. To this end, this work provides a comprehensive analysis of the numerical convergence of unsteady detonation simulations, with focus on isolating the impacts of chemical timescale modifications on convergence characteristics in the context of operator splitting. With the aid of an adaptive mesh refinement based flow solver, the convergence analysis is conducted using two kinetics configurations: (1) a simplified three-step model mechanism, in which chemical timescales in the detonation are modified by adjusting activation energies, and (2) a detailed hydrogen mechanism, in which chemical timescales are adjusted through ambient pressure modifications. The convergence of unsteady self-sustained detonations in one-dimensional channels is then analyzed with reference to steady-state theoretical baseline solutions using these mechanisms. The goal of the analysis is to provide a detailed comparison of the effects of grid resolution on both macroscopic (peak pressures and detonation wave speeds) and microscopic (detonation wave structure) quantities of interest, drawing connections between the deviations from steady-state baselines and minimum chemical timescales. This work uncovers resolution-dependent unsteady detonation regimes, and highlights the important role played by not only the chemical timescales, but also the ratio between chemical timescale and induction time in the detonation wave structure on simulation convergence properties.

In this paper, we study differentially private (DP) algorithms for computing the geometric median (GM) of a dataset: Given $n$ points, $x_1,\dots,x_n$ in $\mathbb{R}^d$, the goal is to find a point $\theta$ that minimizes the sum of the Euclidean distances to these points, i.e., $\sum_{i=1}^{n} \|\theta - x_i\|_2$. Off-the-shelf methods, such as DP-GD, require strong a priori knowledge locating the data within a ball of radius $R$, and the excess risk of the algorithm depends linearly on $R$. In this paper, we ask: can we design an efficient and private algorithm with an excess error guarantee that scales with the (unknown) radius containing the majority of the datapoints? Our main contribution is a pair of polynomial-time DP algorithms for the task of private GM with an excess error guarantee that scales with the effective diameter of the datapoints. Additionally, we propose an inefficient algorithm based on the inverse smooth sensitivity mechanism, which satisfies the more restrictive notion of pure DP. We complement our results with a lower bound and demonstrate the optimality of our polynomial-time algorithms in terms of sample complexity., Comment: 36 pages

We study person-level differentially private (DP) mean estimation in the case where each person holds multiple samples. DP here requires the usual notion of distributional stability when $\textit{all}$ of a person's datapoints can be modified. Informally, if $n$ people each have $m$ samples from an unknown $d$-dimensional distribution with bounded $k$-th moments, we show that \[n = \tilde \Theta\left(\frac{d}{\alpha^2 m} + \frac{d}{\alpha m^{1/2} \varepsilon} + \frac{d}{\alpha^{k/(k-1)} m \varepsilon} + \frac{d}{\varepsilon}\right)\] people are necessary and sufficient to estimate the mean up to distance $\alpha$ in $\ell_2$-norm under $\varepsilon$-differential privacy (and its common relaxations). In the multivariate setting, we give computationally efficient algorithms under approximate-DP and computationally inefficient algorithms under pure DP, and our nearly matching lower bounds hold for the most permissive case of approximate DP. Our computationally efficient estimators are based on the standard clip-and-noise framework, but the analysis for our setting requires both new algorithmic techniques and new analyses. In particular, our new bounds on the tails of sums of independent, vector-valued, bounded-moments random variables may be of interest., Comment: 72 pages, 3 figures

Biosignature detection in the atmospheres of Earth-like exoplanets is one of the most significant and ambitious goals for astronomy, astrobiology, and humanity. Molecular oxygen is among the strongest indicators of life on Earth, but it will be extremely difficult to detect via transmission spectroscopy. We used the Bioverse statistical framework to assess the ability to probe Earth-like O$_{\mathrm{2}}$ levels on hypothetical nearby habitable zone exoplanets (EECs) using direct imaging and high-resolution spectroscopy on the Giant Magellan Telescope (GMT) and the Extremely Large Telescope (ELT). We found that O$_{\mathrm{2}}$ could be probed on up to $\sim$5 and $\sim$15 EECs orbiting bright M dwarfs within 20 pc in a 10-year survey on the GMT and ELT, respectively. Earth-like O$_{\mathrm{2}}$ levels could be probed on four known super-Earth candidates, including Proxima Centauri b, within about one week on the ELT and a few months on the GMT. We also assessed the ability of the ELT to test the habitable zone oxygen hypothesis $\unicode{x2013}$ that habitable zone Earth-sized planets are more likely to have O$_{\mathrm{2}}$ $\unicode{x2013}$ within a 10-year survey using Bioverse. Testing this hypothesis requires either $\sim$1/2 of the EECs to have O$_{\mathrm{2}}$ or $\sim$1/3 if $\eta_{\oplus}$ is large. A northern hemisphere large-aperture telescope, such as the Thirty Meter Telescope (TMT), would expand the target star pool by about 25%, reduce the time to probe biosignatures on individual targets, and provide an additional independent check on potential biosignature detections., Comment: 19 pages, 8 figures, 4 tables. Under review, comments are welcome

State-of-the-art Vision-Language Models (VLMs) ground the vision and the language modality primarily via projecting the vision tokens from the encoder to language-like tokens, which are directly fed to the Large Language Model (LLM) decoder. While these models have shown unprecedented performance in many downstream zero-shot tasks (eg image captioning, question answers, etc), still little emphasis has been put on transferring one of the core LLM capability of In-Context Learning (ICL). ICL is the ability of a model to reason about a downstream task with a few examples demonstrations embedded in the prompt. In this work, through extensive evaluations, we find that the state-of-the-art VLMs somewhat lack the ability to follow ICL instructions. In particular, we discover that even models that underwent large-scale mixed modality pre-training and were implicitly guided to make use of interleaved image and text information (intended to consume helpful context from multiple images) under-perform when prompted with few-shot demonstrations (in an ICL way), likely due to their lack of direct ICL instruction tuning. To enhance the ICL abilities of the present VLM, we propose a simple yet surprisingly effective multi-turn curriculum-based learning methodology with effective data mixes, leading up to a significant 21.03% (and 11.3% on average) ICL performance boost over the strongest VLM baselines and a variety of ICL benchmarks. Furthermore, we also contribute new benchmarks for ICL evaluation in VLMs and discuss their advantages over the prior art.

Newborn animals have advanced perceptual skills at birth, but the nature of this initial knowledge is unknown. Is initial knowledge flexible, continuously adapting to the statistics of experience? Or can initial knowledge be rigid and robust to change, even in the face of opposing evidence? We address this question through controlled-rearing experiments on newborn chicks. First, we reared chicks in an impoverished virtual world, where objects never occluded one another, and found that chicks still succeed on object permanence tasks. Second, we reared chicks in a virtual world in which objects teleported from one location to another while out of view: an unnatural event that violates the continuity of object motion. Despite seeing thousands of these violations of object permanence, and not a single non-violation, the chicks behaved as if object permanence were true, exhibiting the same behavior as chicks reared with natural object permanence events. We conclude that object permanence develops prenatally and is robust to change from opposing evidence., Comment: 10 pages, 4 figures

We collected near-infrared spectra of 65 cool stars with the NASA InfraRed Telescope Facility (IRTF) and analyze them to calculate accurate metallicities and stellar parameters. The sample of 55 M dwarfs and 10 K dwarfs includes 25 systems with confirmed planets and 27 systems with planet candidates identified by the K2 and TESS missions. Three of the 25 confirmed planetary systems host multiple confirmed planets and two of the 27 planet candidate systems host multiple planet candidates. Using the new stellar parameters, we re-fit the K2 and TESS light curves to calculate updated planet properties. In general, our updated stellar properties are more precise than those previously reported and our updated planet properties agree well with those in the literature. Lastly, we briefly examine the relationship between stellar mass, stellar metallicity, and planetary system properties for targets in our sample and for previously characterized planet-hosting low-mass stars. We provide our spectra, stellar parameters, and new planetary fits to the community, expanding the sample available with which to investigate correlations between stellar and planetary properties for low-mass stars., Comment: 4 tables, 6 figures

We provide a simple and flexible framework for designing differentially private algorithms to find approximate stationary points of non-convex loss functions. Our framework is based on using a private approximate risk minimizer to "warm start" another private algorithm for finding stationary points. We use this framework to obtain improved, and sometimes optimal, rates for several classes of non-convex loss functions. First, we obtain improved rates for finding stationary points of smooth non-convex empirical loss functions. Second, we specialize to quasar-convex functions, which generalize star-convex functions and arise in learning dynamical systems and training some neural nets. We achieve the optimal rate for this class. Third, we give an optimal algorithm for finding stationary points of functions satisfying the Kurdyka-Lojasiewicz (KL) condition. For example, over-parameterized neural networks often satisfy this condition. Fourth, we provide new state-of-the-art rates for stationary points of non-convex population loss functions. Fifth, we obtain improved rates for non-convex generalized linear models. A modification of our algorithm achieves nearly the same rates for second-order stationary points of functions with Lipschitz Hessian, improving over the previous state-of-the-art for each of the above problems., Comment: ICML 2024

Recently, Heller & Hippke argued that the exomoon candidates Kepler-1625 b-i and Kepler-1708 b-i were allegedly 'refuted'. In this Matters Arising, we address these claims. For Kepler-1625 b, we show that their Hubble light curve is identical to that previously published by the same lead author, in which the moon-like dip was recovered. Indeed, our fits of their data again recover the moon-like dip with improved residuals than that obtained by Heller & Hippke. Their fits therefore appear to have somehow missed this deeper likelihood maximum, as well producing apparently unconverged posteriors. Consequently, their best-fitting moon is the same radius as the planet, Kepler-1625 b; a radically different signal from that which was originally claimed. The authors then inject this solution into the Kepler data and remark, as a point of concern, how retrievals obtain much higher significances than originally reported. However, this issue stems from the injection of a fundamentally different signal. We demonstrate that their Hubble light curve exhibits ~20% higher noise and discards 11% of the useful data, which compromises its ability to recover the subtle signal of Kepler-1625 b-i. For Kepler-1708 b-i it was claimed that the exomoon model's Bayes factor is highly sensitive to detrending choices, yielding reduced evidence with a biweight filter versus the original claim. We use their own i) detrended light curve and ii) biweight filter code to investigate these claims. For both, we recover the original moon signal, to even higher confidence than before. The discrepancy is explained by comparing to their quoted fit metrics, where we again demonstrate that the Heller & Hippke regression definitively missed the deeper likelihood maximum corresponding to Kepler-1708 b-i. We conclude that both candidates remain viable but certainly demand further observations., Comment: Under consideration by Nature Astronomy as Matters Arising

Theory of Mind (ToM), the ability to understand people's mental states, is an essential ingredient for developing machines with human-level social intelligence. Recent machine learning models, particularly large language models, seem to show some aspects of ToM understanding. However, existing ToM benchmarks use unimodal datasets - either video or text. Human ToM, on the other hand, is more than video or text understanding. People can flexibly reason about another person's mind based on conceptual representations (e.g., goals, beliefs, plans) extracted from any available data. To address this, we introduce a multimodal Theory of Mind question answering (MMToM-QA) benchmark. MMToM-QA comprehensively evaluates machine ToM both on multimodal data and on different kinds of unimodal data about a person's activity in a household environment. To engineer multimodal ToM capacity, we propose a novel method, BIP-ALM (Bayesian Inverse Planning Accelerated by Language Models). BIP-ALM extracts unified representations from multimodal data and utilizes language models for scalable Bayesian inverse planning. We conducted a systematic comparison of human performance, BIP-ALM, and state-of-the-art models, including GPT-4. The experiments demonstrate that large language models and large multimodal models still lack robust ToM capacity. BIP-ALM, on the other hand, shows promising results, by leveraging the power of both model-based mental inference and language models., Comment: ACL 2024. 26 pages, 11 figures, 7 tables

Metalearning and multitask learning are two frameworks for solving a group of related learning tasks more efficiently than we could hope to solve each of the individual tasks on their own. In multitask learning, we are given a fixed set of related learning tasks and need to output one accurate model per task, whereas in metalearning we are given tasks that are drawn i.i.d. from a metadistribution and need to output some common information that can be easily specialized to new tasks from the metadistribution. We consider a binary classification setting where tasks are related by a shared representation, that is, every task $P$ can be solved by a classifier of the form $f_{P} \circ h$ where $h \in H$ is a map from features to a representation space that is shared across tasks, and $f_{P} \in F$ is a task-specific classifier from the representation space to labels. The main question we ask is how much data do we need to metalearn a good representation? Here, the amount of data is measured in terms of the number of tasks $t$ that we need to see and the number of samples $n$ per task. We focus on the regime where $n$ is extremely small. Our main result shows that, in a distribution-free setting where the feature vectors are in $\mathbb{R}^d$, the representation is a linear map from $\mathbb{R}^d \to \mathbb{R}^k$, and the task-specific classifiers are halfspaces in $\mathbb{R}^k$, we can metalearn a representation with error $\varepsilon$ using $n = k+2$ samples per task, and $d \cdot (1/\varepsilon)^{O(k)}$ tasks. Learning with so few samples per task is remarkable because metalearning would be impossible with $k+1$ samples per task, and because we cannot even hope to learn an accurate task-specific classifier with $k+2$ samples per task. Our work also yields a characterization of distribution-free multitask learning and reductions between meta and multitask learning.

Aspect-based sentiment analysis (ABSA), a nuanced task in text analysis, seeks to discern sentiment orientation linked to specific aspect terms in text. Traditional approaches often overlook or inadequately model the explicit syntactic structures of sentences, crucial for effective aspect term identification and sentiment determination. Addressing this gap, we introduce an innovative model: Syntactic Dependency Enhanced Multi-Task Interaction Architecture (SDEMTIA) for comprehensive ABSA. Our approach innovatively exploits syntactic knowledge (dependency relations and types) using a specialized Syntactic Dependency Embedded Interactive Network (SDEIN). We also incorporate a novel and efficient message-passing mechanism within a multi-task learning framework to bolster learning efficacy. Our extensive experiments on benchmark datasets showcase our model's superiority, significantly surpassing existing methods. Additionally, incorporating BERT as an auxiliary feature extractor further enhances our model's performance.

Large language models (LLMs) trained on huge corpora of text datasets demonstrate intriguing capabilities, achieving state-of-the-art performance on tasks they were not explicitly trained for. The precise nature of LLM capabilities is often mysterious, and different prompts can elicit different capabilities through in-context learning. We propose a framework that enables us to analyze in-context learning dynamics to understand latent concepts underlying LLMs' behavioral patterns. This provides a more nuanced understanding than success-or-failure evaluation benchmarks, but does not require observing internal activations as a mechanistic interpretation of circuits would. Inspired by the cognitive science of human randomness perception, we use random binary sequences as context and study dynamics of in-context learning by manipulating properties of context data, such as sequence length. In the latest GPT-3.5+ models, we find emergent abilities to generate seemingly random numbers and learn basic formal languages, with striking in-context learning dynamics where model outputs transition sharply from seemingly random behaviors to deterministic repetition.

Reliable detections of Earth-sized planets in the habitable zone remain elusive in the Kepler sample, even for M dwarfs. The Kepler sample was once thought to contain a considerable number of M dwarf stars ($T_\mathrm{eff} < 4000$ K), which hosted enough Earth-sized ($[0.5,1.5]$ R$_\oplus$) planets to estimate their occurrence rate ($\eta_\oplus$) in the habitable zone. However, updated stellar properties from Gaia have shifted many Kepler stars to earlier spectral type classifications, with most stars (and their planets) now measured to be larger and hotter than previously believed. Today, only one partially-reliable Earth-sized candidate remains in the optimistic habitable zone, and zero in the conservative zone. Here we performed a new investigation of Kepler's Earth-sized planets orbiting M dwarf stars, using occurrence rate models with considerations of updated parameters and candidate reliability. Extrapolating our models to low instellations, we found an occurrence rate of $\eta_\oplus={8.58}_{-8.22}^{+17.94}\%$ for the conservative habitable zone (and ${14.22}_{-12.71}^{+24.96}\%$ for the optimistic), consistent with previous works when considering the large uncertainties. Comparing these estimates to those from similarly comprehensive studies of Sun-like stars, we found that the current Kepler sample does not offer evidence to support an increase in $\eta_\oplus$ from FGK to M stars. While the Kepler sample is too sparse to resolve an occurrence trend between early and mid-to-late M dwarfs for Earth-sized planets, studies including larger planets and/or data from the K2 and TESS missions are well-suited to this task., Comment: 22 pages, 11 figures, 2 tables; Accepted for publication in AJ

Over the last decade, precise exoplanet transmission spectroscopy has revealed the atmospheres of dozens of exoplanets, driven largely by observatories like the Hubble Space Telescope. One major discovery has been the ubiquity of atmospheric aerosols, often blocking access to exoplanet chemical inventories. Tentative trends have been identified, showing that the clarity of planetary atmospheres may depend on equilibrium temperature. Previous work has often grouped dissimilar planets together in order to increase the statistical power of any trends, but it remains unclear from observed transmission spectra whether these planets exhibit the same atmospheric physics and chemistry. We present a re-analysis of a smaller, more physically similar sample of 15 exo-Neptune transmission spectra across a wide range of temperatures (200-1000 K). Using condensation cloud and hydrocarbon haze models, we find that the exo-Neptune population is best described by low cloud sedimentation efficiency ($\mathrm{f_{sed}}\sim0.1$) and high metallicity ($100\times$ Solar). There is an intrinsic scatter of $\sim0.5$ scale height, perhaps evidence of stochasticity in these planets' formation processes. Observers should expect significant attenuation in transmission spectra of Neptune-size exoplanets, up to 6 scale heights for equilibrium temperatures between 500 and 800 K. With JWST's greater wavelength sensitivity, colder (<500 K) planets should be high-priority targets given their clearer atmospheres, and the need to distinguish between the "super-puffs" and more typical gas-dominated planets., Comment: Published in ApJL. 11 pages, 3 figures, 4 tables

The integration of machine learning (ML) in numerous critical applications introduces a range of privacy concerns for individuals who provide their datasets for model training. One such privacy risk is Membership Inference (MI), in which an attacker seeks to determine whether a particular data sample was included in the training dataset of a model. Current state-of-the-art MI attacks capitalize on access to the model's predicted confidence scores to successfully perform membership inference, and employ data poisoning to further enhance their effectiveness. In this work, we focus on the less explored and more realistic label-only setting, where the model provides only the predicted label on a queried sample. We show that existing label-only MI attacks are ineffective at inferring membership in the low False Positive Rate (FPR) regime. To address this challenge, we propose a new attack Chameleon that leverages a novel adaptive data poisoning strategy and an efficient query selection method to achieve significantly more accurate membership inference than existing label-only attacks, especially at low FPRs., Comment: To appear at International Conference on Learning Representations (ICLR) 2024

Recent work on the characterization of small exoplanets has allowed us to accumulate growing evidence that the sub-Neptunes with radii greater than $\sim2.5\,R_\oplus$ often host H$_2$/He-dominated atmospheres both from measurements of their low bulk densities and direct detections of their low mean-molecular-mass atmospheres. However, the smaller sub-Neptunes in the 1.5-2.2 R$_\oplus$ size regime are much less understood, and often have bulk densities that can be explained either by the H$_2$/He-rich scenario, or by a volatile-dominated composition known as the "water world" scenario. Here, we report the detection of water vapor in the transmission spectrum of the $1.96\pm0.08$ R$_\oplus$ sub-Neptune GJ9827d obtained with the Hubble Space Telescope. We observed 11 HST/WFC3 transits of GJ9827d and find an absorption feature at 1.4$\mu$m in its transit spectrum, which is best explained (at 3.39$\sigma$) by the presence of water in GJ9827d's atmosphere. We further show that this feature cannot be caused by unnoculted star spots during the transits by combining an analysis of the K2 photometry and transit light-source effect retrievals. We reveal that the water absorption feature can be similarly well explained by a small amount of water vapor in a cloudy H$_2$/He atmosphere, or by a water vapor envelope on GJ9827d. Given that recent studies have inferred an important mass-loss rate ($>0.5\,$M$_\oplus$/Gyr) for GJ9827d making it unlikely to retain a H-dominated envelope, our findings highlight GJ9827d as a promising water world candidate that could host a volatile-dominated atmosphere. This water detection also makes GJ9827d the smallest exoplanet with an atmospheric molecular detection to date., Comment: Published in ApJL, 11 pages, 6 figures

Long-term magma ocean phases on rocky exoplanets orbiting closer to their star than the runaway greenhouse threshold - the inner edge of the classical habitable zone - may offer insights into the physical and chemical processes that distinguish potentially habitable worlds from others. Thermal stratification of runaway planets is expected to significantly inflate their atmospheres, potentially providing observational access to the runaway greenhouse transition in the form of a "habitable zone inner edge discontinuity" in radius-density space. Here, we use Bioverse, a statistical framework combining contextual information from the overall planet population with a survey simulator, to assess the ability of ground- and space-based telescopes to test this hypothesis. We find that the demographic imprint of the runaway greenhouse transition is likely detectable with high-precision transit photometry for sample sizes $\gtrsim 100$ planets if at least ~10 % of those orbiting closer than the habitable zone inner edge harbor runaway climates. Our survey simulations suggest that in the near future, ESA's PLATO mission will be the most promising survey to probe the habitable zone inner edge discontinuity. We determine survey strategies that maximize the diagnostic power of the obtained data and identify as key mission design drivers: 1. A follow-up campaign of planetary mass measurements and 2. The fraction of low-mass stars in the target sample. Observational constraints on the runaway greenhouse transition will provide crucial insights into the distribution of atmospheric volatiles among rocky exoplanets, which may help to identify the nearest potentially habitable worlds., Comment: For a video abstract, see https://youtu.be/acgKcdTTv9c. All source code is available at https://github.com/matiscke/hz-inner-edge-discontinuity. 29 pages, 12 figures, 1 table. Typos corrected, references added. Published in The Planetary Science Journal

The demographics of young exoplanets can shed light onto their formation and evolution processes. Exoplanet properties are derived from the properties of their host stars. As such, it is important to accurately characterize the host stars since any systematic biases in their derivation can negatively impact the derivation of planetary properties. Here, we present a uniform catalog of photometrically-derived stellar effective temperatures, luminosities, radii, and masses for 4,865 young (<1 Gyr) stars in 31 nearby clusters and moving groups within 200 pc. We compared our photometrically-derived properties to a subset of those derived from spectra, and found them to be in good agreement. We also investigated the effect of stellar properties on the detection efficiency of transiting short-period young planets with TESS as calculated in Fernandes et al. 2022, and found an overall increase in the detection efficiency when the new photometrically derived properties were taken into account. Most notably, there is a 1.5 times increase in the detection efficiencies for sub-Neptunes/Neptunes (1.8-6 Re) implying that, for our sample of young stars, better characterization of host star properties can lead to the recovery of more small transiting planets. Our homogeneously derived catalog of updated stellar properties, along with a larger unbiased stellar sample and more detections of young planets, will be a crucial input to the accurate estimation of the occurrence rates of young short-period planets., Comment: 16 pages, 5 Figures, 3 Tables. Revised and resubmitted to AJ after a favorable referee report. Co-First Authors