arXiv subjects

Daniel Schwartz

Publications and source records attributed to Daniel Schwartz.

10 recordsLinked to original sources

arXiv2026

Resource-Efficient Gesture Recognition through Convexified Attention

Wearable e-textile interfaces require gesture recognition capabilities but face severe constraints in power consumption, computational capacity, and form factor that make traditional deep learning impractical. While lightweight architectures like MobileNet improve efficiency, they still demand thousands of parameters, limiting deployment on textile-integrated platforms. We introduce a convexified attention mechanism for wearable applications that dynamically weights features while preserving convexity through nonexpansive simplex projection and convex loss functions. Unlike conventional attention mechanisms using non-convex softmax operations, our approach employs Euclidean projection onto the probability simplex combined with multi-class hinge loss, ensuring global convergence guarantees. Implemented on a textile-based capacitive sensor with four connection points, our approach achieves 100.00\% accuracy on tap gestures and 100.00\% on swipe gestures -- consistent across 10-fold cross-validation and held-out test evaluation -- while requiring only 120--360 parameters, a 97\% reduction compared to conventional approaches. With sub-millisecond inference times (290--296$\mu$s) and minimal storage requirements ($<$7KB), our method enables gesture interfaces directly within e-textiles without external processing. Our evaluation, conducted in controlled laboratory conditions with a single-user dataset, demonstrates feasibility for basic gesture interactions. Real-world deployment would require validation across multiple users, environmental conditions, and more complex gesture vocabularies. These results demonstrate how convex optimization can enable efficient on-device machine learning for textile interfaces.

cs.LG↗

arXiv2025

Determining the spatial origin of X-ray and optical emission in the $z = 3.1$ strongly lensed radio-quiet quasar GraL J065904.1+162909 to hundreds of parsecs

We perform milliarcsecond X-ray astrometry of the quadruply lensed radio-quiet quasar GraL J065904.1+162909 (J0659). This $z = 3.083$ quasar is lensed into four images and was discovered with the second Data Release of the $Gaia$ Space Observatory ($Gaia$ DR2). Our J0659 study exploits strong gravitational lenses as high resolution telescopes. This technique shows promise to elucidate the origin of optical and X-ray emission in distant lensed quasars at spatial scales beyond the reach of current instruments. In our study, we use $Gaia$ DR3 and $HST$ observations of J0659 to infer a mass model for the deflector. Our model reproduces the $Gaia$ DR3 quasar lensed image positions to one milliarcsecond and determines the position of the optical source in J0659 to within this precision. Next, we analyze $Chandra$ observations of J0659 and conduct a Bayesian test evaluating whether the X-ray emission region coincides with the optical source. We then constrain the origin of the X-ray emission to within a $0.''020 \times 0.''010$ ellipse centered $0.''014$ away from the optical source at the $1\sigma$ level. We demonstrate that our approach can be extended to pinpoint the distinct origins of the soft and hard X-ray emission regions in lensed quasars. We discuss the potential of upcoming broadband and spectrally resolved X-ray astrometric studies to probe complex quasar morphology and AGN multiplicity at sub-kiloparsec scales otherwise inaccessible at high redshifts.

astro-ph.HE↗

arXiv2025

Zero-knowledge LLM hallucination detection and mitigation through fine-grained cross-model consistency

Large language models (LLMs) have demonstrated impressive capabilities across diverse tasks, but they remain susceptible to hallucinations--generating content that appears plausible but contains factual inaccuracies. We present Finch-Zk, a black-box framework that leverages fine-grained cross-model consistency to detect and mitigate hallucinations in LLM outputs without requiring external knowledge sources. Finch-Zk introduces two key innovations: 1) a cross-model consistency checking strategy that reveals fine-grained inaccuracies by comparing responses generated by diverse models from semantically-equivalent prompts, and 2) a targeted mitigation technique that applies precise corrections to problematic segments while preserving accurate content. Experiments on the FELM dataset show Finch-Zk improves hallucination detection F1 scores by 6-39\% compared to existing approaches. For mitigation, Finch-Zk achieves up to 9 absolute percentage points improvement in answer accuracy on the GPQA-diamond dataset when applied to state-of-the-art models like Llama 4 Maverick and Claude 4 Sonnet. Extensive evaluation on multiple datasets demonstrates that Finch-Zk provides a practical, deployment-ready safeguard for enhancing factual reliability in production LLM systems.

cs.CL↗

arXiv2025

Milli-arcsec X-ray positions and X-ray varstrometry for the strongly lensed AGN HE 0435-1223

Active galactic nuclei (AGN) are some of the most powerful objects in the Universe. For this reason, they can be observed up to high redshifts (z), giving valuable insights into the evolution of our Universe. However, high-z AGN are too distant to be spatially resolved with current or upcoming X-ray facilities. In this paper we show how we can exploit gravitationally lensed AGN to significantly increase spatial resolution even at high-z. We combine astrometric data from Gaia DR3 with imaging from the Chandra X-ray Observatory of the quadruply-lensed quasar HE 0435--1223 to measure for the first time possible offsets between the optical and the X-ray emissions. We measure the X-ray source position for HE 0435-1223 within a 1$\sigma$ quasi-elliptical region of 0.5 x 1.3 milli-arcsecond (mas), about 150 pc$^2$ at the redshift of the source (z=1.689). We find evidence for the X-ray emission being offset by a projected 3 mas from the Gaia (optical) emission. The positional offset is most likely associated to a portion of the X-ray emission arising from an X-ray jet or outflow. We also discuss how this method can be used to indicate the presence of a binary/offset AGN system.

astro-ph.HE↗

arXiv2025

Graph of Attacks with Pruning: Optimizing Stealthy Jailbreak Prompt Generation for Enhanced LLM Content Moderation

As large language models (LLMs) become increasingly prevalent, ensuring their robustness against adversarial misuse is crucial. This paper introduces the GAP (Graph of Attacks with Pruning) framework, an advanced approach for generating stealthy jailbreak prompts to evaluate and enhance LLM safeguards. GAP addresses limitations in existing tree-based LLM jailbreak methods by implementing an interconnected graph structure that enables knowledge sharing across attack paths. Our experimental evaluation demonstrates GAP's superiority over existing techniques, achieving a 20.8% increase in attack success rates while reducing query costs by 62.7%. GAP consistently outperforms state-of-the-art methods for attacking both open and closed LLMs, with attack success rates of >96%. Additionally, we present specialized variants like GAP-Auto for automated seed generation and GAP-VLM for multimodal attacks. GAP-generated prompts prove highly effective in improving content moderation systems, increasing true positive detection rates by 108.5% and accuracy by 183.6% when used for fine-tuning. Our implementation is available at https://github.com/dsbuddy/GAP-LLM-Safety.

cs.CR↗

arXiv2023

Harmonized Estimation of Subgroup-Specific Treatment Effects in Randomized Trials: The Use of External Control Data

Subgroup analyses of randomized controlled trials (RCTs) constitute an important component of the drug development process in precision medicine. In particular, subgroup analyses of early-stage trials often influence the design and eligibility criteria of subsequent confirmatory trials and ultimately influence which subpopulations will receive the treatment after regulatory approval. However, subgroup analyses are often complicated by small sample sizes, which leads to substantial uncertainty about subgroup-specific treatment effects. We explore the use of external control (EC) data to augment RCT subgroup analyses. We define and discuss harmonized estimators of subpopulation-specific treatment effects that leverage EC data. Our approach can be used to modify any subgroup-specific treatment effect estimates that are obtained by combining RCT and EC data, such as linear regression. We alter these subgroup-specific estimates to make them coherent with a robust estimate of the average effect in the randomized population based only on RCT data. The weighted average of the resulting subgroup-specific harmonized estimates matches the RCT-only estimate of the overall effect in the randomized population. We discuss the proposed harmonized estimators through analytic results and simulations, and investigate standard performance metrics. The method is illustrated with a case study in oncology.

stat.ME↗

arXiv2022

EvoSTS Forecasting: Evolutionary Sparse Time-Series Forecasting

In this work, we highlight our novel evolutionary sparse time-series forecasting algorithm also known as EvoSTS. The algorithm attempts to evolutionary prioritize weights of Long Short-Term Memory (LSTM) Network that best minimize the reconstruction loss of a predicted signal using a learned sparse coded dictionary. In each generation of our evolutionary algorithm, a set number of children with the same initial weights are spawned. Each child undergoes a training step and adjusts their weights on the same data. Due to stochastic back-propagation, the set of children has a variety of weights with different levels of performance. The weights that best minimize the reconstruction loss with a given signal dictionary are passed to the next generation. The predictions from the best-performing weights of the first and last generation are compared. We found improvements while comparing the weights of these two generations. However, due to several confounding parameters and hyperparameter limitations, some of the weights had negligible improvements. To the best of our knowledge, this is the first attempt to use sparse coding in this way to optimize time series forecasting model weights, such as those of an LSTM network.

cs.NE↗

arXiv2021

Resolving Complex Inner X-ray Structure of the Gravitationaly Lensed AGN MGB2016+112

We use a Chandra X-ray observation of the gravitationally lensed system MGB2016+112 at z=3.273 to elucidate presence of at least two X-ray sources. We find that these sources are consistent with the VLBI components measured by \citet{Spingola19}, which are separated by $\sim 200$ pc. Their intrinsic 0.5 -- 7 keV source frame luminosities are 2.6$\times$10$^{43}$ and 4.2$\times$10$^{44}$ erg s$^{-1}$. Most likely this system contains a dual active galactic nucleus (AGN), but we possibly are detecting an AGN plus a pc-scale X-ray jet, the latter lying in a region at very high magnification. The quadruply lensed X-ray source is within $\pm$40 pc (1$σ$) of its VLBI counterpart. Using a gravitational lens as a telescope, and a novel statistical application, we have achieved unprecedented accuracy for measuring metric distances at such large redshifts in X-ray astronomy, which is tens of mas if the source is located close to the caustics, while it is of hundreds of mas if the source is in a region at lower amplification. The present demonstration of this approach has implications for future X-ray investigations of large numbers of lensed systems.

astro-ph.HE↗

arXiv2021

Towards Searching Efficient and Accurate Neural Network Architectures in Binary Classification Problems

In recent years, deep neural networks have had great success in machine learning and pattern recognition. Architecture size for a neural network contributes significantly to the success of any neural network. In this study, we optimize the selection process by investigating different search algorithms to find a neural network architecture size that yields the highest accuracy. We apply binary search on a very well-defined binary classification network search space and compare the results to those of linear search. We also propose how to relax some of the assumptions regarding the dataset so that our solution can be generalized to any binary classification problem. We report a 100-fold running time improvement over the naive linear search when we apply the binary search method to our datasets in order to find the best architecture candidate. By finding the optimal architecture size for any binary classification problem quickly, we hope that our research contributes to discovering intelligent algorithms for optimizing architecture size selection in machine learning.

cs.LG↗

arXiv2020

Two Candidate High-Redshift X-ray Jets Without Coincident Radio Jets

We report the detection of extended X-ray emission from two high-redshift radio quasars. These quasars, J1405+0415 at $z$=3.208 and J1610+1811 at $z$=3.118, were observed in a Chandra snapshot survey selected from a complete sample of the radio-brightest quasars in the overlap area of the VLA-FIRST radio survey and the Sloan Digital Sky Survey. The extended X-ray emission is located along the line connecting the core to a radio knot or hotspot, favoring the interpretation of X-ray jets. The inferred rest frame jet X-ray luminosities from 2--30 keV would be of order 10$^{45}$ erg~s$^{-1}$ if emitted isotropically and without relativistic beaming. In the scenario of inverse Compton scattering of the cosmic microwave background (CMB), X-ray jets without a coincident radio counterpart may be common, and should be readily detectable to redshifts even beyond 3.2 due to the (1+$z$)$^4$ increase of the CMB energy density compensating for the (1+$z$)$^{-4}$ cosmological diminution of surface brightness. If these can be X-ray confirmed, they would be the second and third examples of quasar X-ray jets without detection of underlying continuous radio jets.

astro-ph.HE↗