Search

Malicious Python packages make software supply chains vulnerable by exploiting trust in open-source repositories like Python Package Index (PyPI). Lack of real-time behavioral monitoring makes metadata inspection and static code analysis inadequate against advanced attack strategies such as typosquatting, covert remote access activation, and dynamic payload generation. To address these challenges, we introduce DySec, a machine learning (ML)-based dynamic analysis framework for PyPI that uses eBPF kernel and user-level probes to monitor behaviors during package installation. By capturing 36 real-time features-including system calls, network traffic, resource usage, directory access, and installation patterns-DySec detects threats like typosquatting, covert remote access activation, dynamic payload generation, and multiphase attack malware. We developed a comprehensive dataset of 14,271 Python packages, including 7,127 malicious sample traces, by executing them in a controlled isolated environment. Experimental results demonstrate that DySec achieves a 95.99\% detection accuracy with a latency of <0.5s, reducing false negatives by 78.65\% compared to static analysis and 82.24\% compared to metadata analysis. During the evaluation, DySec flagged 11 packages that PyPI classified as benign. A manual analysis, including installation behavior inspection, confirmed six of them as malicious. These findings were reported to PyPI maintainers, resulting in the removal of four packages. DySec bridges the gap between reactive traditional methods and proactive, scalable threat mitigation in open-source ecosystems by uniquely detecting malicious install-time behaviors.

Trapped-ion systems are a leading platform for quantum computing. The M{\o}lmer-S{\o}rensen (MS) gate is a widely used method for implementing controlled interactions in multipartite systems. However, due to unavoidable interactions with the environment, quantum states undergo non-unitary evolution, leading to significant deviations from ideal dynamics. Common techniques such as Quantum Process Tomography (QPT) and Bell State Tomography (BST) are typically employed to evaluate MS gate performance and to characterize noise in the system. In this letter, we propose leveraging the geometric phase as a tool for performance assessment and noise identification in the MS gate. Our findings indicate that the geometric phase is particularly sensitive to environmental noise occurring around twice the clock pulse time. Given that geometric phase measurements do not require full-state tomography, this approach offers a practical and experimentally feasible method to detect entanglement and classify the nature of noise affecting the system., Comment: Comments are welcome

Knowledge distillation is used, in generative language modeling, to train a smaller student model using the help of a larger teacher model, resulting in improved capabilities for the student model. In this paper, we formulate a more general framework for knowledge distillation where the student learns from the teacher during training, and also learns to ask for the teacher's help at test-time following rules specifying test-time restrictions. Towards this, we first formulate knowledge distillation as an entropy-regularized value optimization problem. Adopting Path Consistency Learning to solve this, leads to a new knowledge distillation algorithm using on-policy and off-policy demonstrations. We extend this using constrained reinforcement learning to a framework that incorporates the use of the teacher model as a test-time reference, within constraints. In this situation, akin to a human learner, the model needs to learn not only the learning material, but also the relative difficulty of different sections to prioritize for seeking teacher help. We examine the efficacy of our method through experiments in translation and summarization tasks, observing trends in accuracy and teacher use, noting that our approach unlocks operating points not available to the popular Speculative Decoding approach.

The blue supergiant (BSG) domain contains a large variety of stars whose past and future evolutionary paths are still highly uncertain. Since binary interaction plays a crucial role in the fate of massive stars, investigating the multiplicity among BSGs helps shed light on the fate of such objects. We aim to estimate the binary fraction of a large sample of BSGs in the Small Magellanic Cloud within the Binarity at LOw Metallicity (BLOeM) survey. In total, we selected 262 targets with spectral types B0-B3 and luminosity classes I-II. This work is based on spectroscopic data collected by the GIRAFFE instrument, mounted on the Very Large Telescope, which gathered nine epochs over three months. Our spectroscopic analysis for each target includes the individual and peak-to-peak radial velocity measurements, an investigation of the line profile variability, and a periodogram analysis to search for possible short- and long-period binaries. By applying a 20 km s$^{-1}$ threshold on the peak-to-peak radial velocities above which we would consider the star to be binary, the resulting observed spectroscopic binary fraction for our BSG sample is 23 $\pm$ 3$\%$. In addition, we derived reliable orbital periods for 41 spectroscopic binaries and potential binary candidates, among which there are 17 eclipsing binaries, including 20 SB1 and SB2 systems with periods of less than 10 days. We reported a significant drop in the binary fraction of BSGs with spectral types later than B2 and effective temperatures less than 18 kK, which could indicate the end of the main sequence phase in this temperature regime. We found no metallicity dependence in the binary fraction of BSGs, compared to existing spectroscopic surveys of the Galaxy and Large Magellanic Cloud., Comment: 18 pages, 14 figures, accepted for publication in Astronomy & Astrophysics

Genetic data collection has become ubiquitous today. The ability to meaningfully interpret genetic data has motivated its widespread use, providing crucial insights into human health and ancestry while driving important public health initiatives. Easy access to genetic testing has fueled a rapid expansion of recreational direct-to-consumer offerings. However, the growth of genetic datasets and their applications has created significant privacy and discrimination risks, as our understanding of the scientific basis for genetic traits continues to evolve. In this paper, we organize the uses of genetic data along four distinct "pillars": clinical practice, research, forensic and government use, and recreational use. Using our scientific understanding of genetics, genetic inference methods and their associated risks, and current public protections, we build a risk assessment framework that identifies key values that any governance system must preserve. We analyze case studies using this framework to assess how well existing regulatory frameworks preserve desired values. Our investigation reveals critical gaps in these frameworks and identifies specific threats to privacy and personal liberties, particularly through genetic discrimination. We propose comprehensive policy reforms to: (1) update the legal definition of genetic data to protect against modern technological capabilities, (2) expand the Genetic Information Nondiscrimination Act (GINA) to cover currently unprotected domains, and (3) establish a unified regulatory framework under a single governing body to oversee all applications of genetic data. We conclude with three open questions about genetic data: the challenges posed by its relational nature, including consent for relatives and minors; the complexities of international data transfer; and its potential integration into large language models., Comment: Submitted to FACCT 2025

Global leaders and policymakers are unified in their unequivocal commitment to decarbonization efforts in support of Net-Zero agreements. District Heating Systems (DHS), while contributing to carbon emissions due to the continued reliance on fossil fuels for heat production, are embracing more sustainable practices albeit with some sense of vulnerability as it could constrain their ability to adapt to dynamic demand and production scenarios. As demographic demands grow and renewables become the central strategy in decarbonizing the heating sector, the need for accurate demand forecasting has intensified. Advances in digitization have paved the way for Machine Learning (ML) based solutions to become the industry standard for modeling complex time series patterns. In this paper, we focus on building a Deep Learning (DL) model that uses deconstructed components of independent and dependent variables that affect heat demand as features to perform multi-step ahead forecasting of head demand. The model represents the input features in a time-frequency space and uses an attention mechanism to generate accurate forecasts. The proposed method is evaluated on a real-world dataset and the forecasting performance is assessed against LSTM and CNN-based forecasting models. Across different supply zones, the attention-based models outperforms the baselines quantitatively and qualitatively, with an Mean Absolute Error (MAE) of 0.105 with a standard deviation of 0.06kW h and a Mean Absolute Percentage Error (MAPE) of 5.4% with a standard deviation of 2.8%, in comparison the second best model with a MAE of 0.10 with a standard deviation of 0.06kW h and a MAPE of 5.6% with a standard deviation of 3%.

Rapidly rotating classical OBe stars have been proposed as the products of binary interactions, and the fraction of Be stars with compact companions implies that at least some are. However, to constrain the interaction physics spinning up the OBe stars, a large sample of homogeneously analysed OBe stars with well-determined binary characteristics and orbital parameters are required. We investigate the multiplicity properties of a sample of 18 Oe, 62 Be, and two Of?p stars observed within the BLOeM survey in the Small Magellanic Cloud. We analyse the first nine epochs of spectroscopic observations obtained over approximately three months in 2023. Radial velocities (RVs) of all stars are measured. Applying commonly-used binarity criteria we classify objects as binaries, binary candidates, and apparently single (RV stable) objects. We further inspect the spectra for double-lined spectroscopic binaries and cross-match with catalogues of X-ray sources and photometric binaries. We classify 14 OBe stars as binaries, and an additional 11 as binary candidates. The two Of?p stars are apparently single. Two more objects are most likely currently interacting binaries. Without those, the observed binary fraction for the OBe sample (78 stars) is f_OBe_obs=0.18+/-0.04 (f_obs_cand=0.32+/-0.05 including candidates). This fraction is less than half of that measured for OB stars in BLOeM. Combined with the lower fraction of SB2s, this suggests that OBe stars have indeed fundamentally different binary properties than OB stars. We find no evidence for OBe binaries with massive compact companions, in contrast to expectations from binary population synthesis. Our results support the binary scenario as an important formation channel for OBe stars, as post-interaction binaries may have been disrupted or the stripped companions of OBe stars are harder to detect., Comment: 12 pages (+appendix), 7 figures, accepted for publication in A&A

A major challenge in aligning large language models (LLMs) with human preferences is the issue of distribution shift. LLM alignment algorithms rely on static preference datasets, assuming that they accurately represent real-world user preferences. However, user preferences vary significantly across geographical regions, demographics, linguistic patterns, and evolving cultural trends. This preference distribution shift leads to catastrophic alignment failures in many real-world applications. We address this problem using the principled framework of distributionally robust optimization, and develop two novel distributionally robust direct preference optimization (DPO) algorithms, namely, Wasserstein DPO (WDPO) and Kullback-Leibler DPO (KLDPO). We characterize the sample complexity of learning the optimal policy parameters for WDPO and KLDPO. Moreover, we propose scalable gradient descent-style learning algorithms by developing suitable approximations for the challenging minimax loss functions of WDPO and KLDPO. Our empirical experiments demonstrate the superior performance of WDPO and KLDPO in substantially improving the alignment when there is a preference distribution shift.

Recent advances in generative AI have led to large multi-modal models (LMMs) capable of simultaneously processing inputs of various modalities such as text, images, video, and audio. While these models demonstrate impressive capabilities, efficiently serving them in production environments poses significant challenges due to their complex architectures and heterogeneous resource requirements. We present the first comprehensive systems analysis of two prominent LMM architectures, decoder-only and cross-attention, on six representative open-source models. We investigate their multi-stage inference pipelines and resource utilization patterns that lead to unique systems design implications. We also present an in-depth analysis of production LMM inference traces, uncovering unique workload characteristics, including variable, heavy-tailed request distributions, diverse modal combinations, and bursty traffic patterns. Our key findings reveal that different LMM inference stages exhibit highly heterogeneous performance characteristics and resource demands, while concurrent requests across modalities lead to significant performance interference. To address these challenges, we propose a decoupled serving architecture that enables independent resource allocation and adaptive scaling for each stage. We further propose optimizations such as stage colocation to maximize throughput and resource utilization while meeting the latency objectives.

Text-to-SQL systems have unlocked easier access to critical data insights by enabling natural language queries over structured databases. However, deploying such systems in enterprise environments remains challenging due to factors such as large, complex schemas (> 3000 columns), diverse SQL dialects (e.g., BigQuery, Snowflake) and sophisticated query requirements (e.g., transformation, analytics). Current state-of-the-art performance on the Spider 2.0 dataset -- a benchmark built to mimic such complex environments -- remains limited at 20%. Key limitations include inadequate instruction-following, poor long-context comprehension, weak self-refinement, and insufficient dialect-specific knowledge. To address these gaps, we propose ReFoRCE (Self-Refinement Agent with Format Restriction and Column Exploration) which introduces (1) table compression to mitigate long-context limitations (2) format restriction to ensure accurate answer format, and (3) iterative column exploration for enhanced schema understanding. Additionally, it employs self-refinement pipeline consisting of (1) parallelized workflows with voting mechanisms and (2) a Common Table Expression (CTE) based refinement approach to handle unresolved cases. ReFoRCE achieves state-of-the-art results scoring 31.26 on the Spider 2.0-Snow and scoring 30.35 on the Spider 2.0-Lite tasks., Comment: 13 pages, 1 figure

Training a neural network for pixel based classification task using low resolution Landsat images is difficult as the size of the training data is usually small due to less number of available pixels that represent a single class without any mixing with other classes. Due to this scarcity of training data, neural network may not be able to attain expected level of accuracy. This limitation could be overcome using a generative network that aims to generate synthetic data having the same distribution as the sample data with which it is trained. In this work, we have proposed a methodology for improving the performance of ANN classifier to identify built-up pixels in the Landsat$7$ image with the help of developing a simple GAN architecture that could generate synthetic training pixels when trained using original set of sample built-up pixels. To ensure that the marginal and joint distributions of all the bands corresponding to the generated and original set of pixels are indistinguishable, non-parametric Kolmogorov Smirnov Test and Ball Divergence based Equality of Distributions Test have been performed respectively. It has been observed that the overall accuracy and kappa coefficient of the ANN model for built-up classification have continuously improved from $0.9331$ to $0.9983$ and $0.8277$ to $0.9958$ respectively, with the inclusion of generated sets of built-up pixels to the original one.

We address the problem of best policy identification in preference-based reinforcement learning (PbRL), where learning occurs from noisy binary preferences over trajectory pairs rather than explicit numerical rewards. This approach is useful for post-training optimization of generative AI models during multi-turn user interactions, where preference feedback is more robust than handcrafted reward models. In this setting, learning is driven by both an offline preference dataset -- collected from a rater of unknown 'competence' -- and online data collected with pure exploration. Since offline datasets may exhibit out-of-distribution (OOD) biases, principled online data collection is necessary. To address this, we propose Posterior Sampling for Preference Learning ($\mathsf{PSPL}$), a novel algorithm inspired by Top-Two Thompson Sampling, that maintains independent posteriors over the true reward model and transition dynamics. We provide the first theoretical guarantees for PbRL in this setting, establishing an upper bound on the simple Bayesian regret of $\mathsf{PSPL}$. Since the exact algorithm can be computationally impractical, we also provide an approximate version that outperforms existing baselines.

A vast amount of medical knowledge is available for public use through online health forums, and question-answering platforms on social media. The majority of the population in the United States doesn't have the right amount of health literacy to make the best use of that information. Health literacy means the ability to obtain and comprehend the basic health information to make appropriate health decisions. To build the bridge between this gap, organizations advocate adapting this medical knowledge into plain language. Building robust systems to automate the adaptations helps both medical and non-medical professionals best leverage the available information online. The goal of the Plain Language Adaptation of Biomedical Abstracts (PLABA) track is to adapt the biomedical abstracts in English language extracted from PubMed based on the questions asked in MedlinePlus for the general public using plain language at the sentence level. As part of this track, we leveraged the best open-source Large Language Models suitable and fine-tuned for dialog use cases. We compare and present the results for all of our systems and our ranking among the other participants' submissions. Our top performing GPT-4 based model ranked first in the avg. simplicity measure and 3rd on the avg. accuracy measure., Comment: 8 pages, 2 figures, 4 tables

We investigate the problem of strong coordination over a multiple-access channel (MAC) with cribbing encoders. In this configuration, two encoders observe independent and identically distributed (i.i.d.) samples of a source random variable each and encode the inputs to the MAC. The decoder which observes the output of the MAC together with side-information, must generate approximately i.i.d. samples of another random variable which is jointly distributed with the two sources and the side information. We also allow for possible encoder cooperation, where one of the encoders can non-causally crib from the other encoders input. Independent pairwise shared randomness is assumed between each encoder and the decoder at limited rates. Firstly, in the presence of cribbing, we derive an achievable region based on joint source-channel coding. We also prove that in the absence of cribbing, our inner bound is tight for the special case when the MAC is composed of deterministic links, and the sources are conditionally independent given the side information. We then explicitly compute the regions for an example both with and without cribbing between the encoders, and demonstrate that cribbing strictly improves upon the achievable region., Comment: 7 pages, 1 figure. arXiv admin note: substantial text overlap with arXiv:2411.14123

Design space exploration (DSE) plays a crucial role in enabling custom hardware architectures, particularly for emerging applications like AI, where optimized and specialized designs are essential. With the growing complexity of deep neural networks (DNNs) and the introduction of advanced foundational models (FMs), the design space for DNN accelerators is expanding at an exponential rate. Additionally, this space is highly non-uniform and non-convex, making it increasingly difficult to navigate and optimize. Traditional DSE techniques rely on search-based methods, which involve iterative sampling of the design space to find the optimal solution. However, this process is both time-consuming and often fails to converge to the global optima for such design spaces. Recently, AIrchitect v1, the first attempt to address the limitations of search-based techniques, transformed DSE into a constant-time classification problem using recommendation networks. In this work, we propose AIrchitect v2, a more accurate and generalizable learning-based DSE technique applicable to large-scale design spaces that overcomes the shortcomings of earlier approaches. Specifically, we devise an encoder-decoder transformer model that (a) encodes the complex design space into a uniform intermediate representation using contrastive learning and (b) leverages a novel unified representation blending the advantages of classification and regression to effectively explore the large DSE space without sacrificing accuracy. Experimental results evaluated on 10^5 real DNN workloads demonstrate that, on average, AIrchitect v2 outperforms existing techniques by 15% in identifying optimal design points. Furthermore, to demonstrate the generalizability of our method, we evaluate performance on unseen model workloads (LLMs) and attain a 1.7x improvement in inference latency on the identified hardware architecture., Comment: Accepted to DATE 2025

Concept erasure techniques have recently gained significant attention for their potential to remove unwanted concepts from text-to-image models. While these methods often demonstrate success in controlled scenarios, their robustness in real-world applications and readiness for deployment remain uncertain. In this work, we identify a critical gap in evaluating sanitized models, particularly in terms of their performance across various concept dimensions. We systematically investigate the failure modes of current concept erasure techniques, with a focus on visually similar, binomial, and semantically related concepts. We propose that these interconnected relationships give rise to a phenomenon of concept entanglement resulting in ripple effects and degradation in image quality. To facilitate more comprehensive evaluation, we introduce EraseBENCH, a multi-dimensional benchmark designed to assess concept erasure methods with greater depth. Our dataset includes over 100 diverse concepts and more than 1,000 tailored prompts, paired with a comprehensive suite of metrics that together offer a holistic view of erasure efficacy. Our findings reveal that even state-of-the-art techniques struggle with maintaining quality post-erasure, indicating that these approaches are not yet ready for real-world deployment. This highlights the gap in reliability of the concept erasure techniques., Comment: 11 pages main; 9 pages supplemental material

Text-to-image (T2I) generation has made significant advances in recent years, but challenges still remain in the generation of perceptual artifacts, misalignment with complex prompts, and safety. The prevailing approach to address these issues involves collecting human feedback on generated images, training reward models to estimate human feedback, and then fine-tuning T2I models based on the reward models to align them with human preferences. However, while existing reward fine-tuning methods can produce images with higher rewards, they may change model behavior in unexpected ways. For example, fine-tuning for one quality aspect (e.g., safety) may degrade other aspects (e.g., prompt alignment), or may lead to reward hacking (e.g., finding a way to increase rewards without having the intended effect). In this paper, we propose Focus-N-Fix, a region-aware fine-tuning method that trains models to correct only previously problematic image regions. The resulting fine-tuned model generates images with the same high-level structure as the original model but shows significant improvements in regions where the original model was deficient in safety (over-sexualization and violence), plausibility, or other criteria. Our experiments demonstrate that Focus-N-Fix improves these localized quality aspects with little or no degradation to others and typically imperceptible changes in the rest of the image. Disclaimer: This paper contains images that may be overly sexual, violent, offensive, or harmful.