Search

Existing work on improving language model reasoning typically explores a single solution path, which can be prone to errors. Inspired by perspective-taking in social studies, this paper introduces DiPT, a novel approach that complements current reasoning methods by explicitly incorporating diversified viewpoints. This approach allows the model to gain a deeper understanding of the problem's context and identify the most effective solution path during the inference stage. Additionally, it provides a general data-centric AI recipe for augmenting existing data to improve their quality for fine-tuning. Our empirical results demonstrate that DiPT can be flexibly integrated into existing methods that focus on a single reasoning approach, enhancing their reasoning performance and stability when presented with paraphrased problems. Furthermore, we illustrate improved context understanding by maintaining the model's safe outputs against "jailbreaking" prompts intentionally designed to bypass safeguards built into deployed models. Lastly, we show that fine-tuning with data enriched with diverse perspectives can boost the reasoning capabilities of the model compared to fine-tuning with raw data alone., Comment: LLM Reasoning with Perspectives, Preprint

Reinforcement learning from human feedback plays a crucial role in aligning language models towards human preferences, traditionally represented through comparisons between pairs or sets of responses within a given context. While many studies have enhanced algorithmic techniques to optimize learning from such data, this work shifts focus to improving preference learning through a data-centric approach. Specifically, we propose enriching existing preference datasets with machine-generated rationales that explain the reasons behind choices. We develop a simple and principled framework to augment current preference learning methods with rationale information. Our comprehensive analysis highlights how rationales enhance learning efficiency. Extensive experiments reveal that rationale-enriched preference learning offers multiple advantages: it improves data efficiency, accelerates convergence to higher-performing models, and reduces verbosity bias and hallucination. Furthermore, this framework is versatile enough to integrate with various preference optimization algorithms. Overall, our findings highlight the potential of re-imagining data design for preference learning, demonstrating that even freely available machine-generated rationales can significantly boost performance across multiple dimensions. The code repository is available at https: //github.com/reds-lab/preference-learning-with-rationales, Comment: Data-Centric Human Preference Learning with Rationales

Static analysis is sound in theory, but an implementation may unsoundly fail to analyze all of a program's code. Any such omission is a serious threat to the validity of the tool's output. Our work is the first to measure the prevalence of these omissions. Previously, researchers and analysts did not know what is missed by static analysis, what sort of code is missed, or the reasons behind these omissions. To address this gap, we ran 13 static analysis tools and a dynamic analysis on 1000 Android apps. Any method in the dynamic analysis but not in a static analysis is an unsoundness. Our findings include the following. (1) Apps built around external frameworks challenge static analyzers. On average, the 13 static analysis tools failed to capture 61% of the dynamically-executed methods. (2) A high level of precision in call graph construction is a synonym for a high level of unsoundness; (3) No existing approach significantly improves static analysis soundness. This includes those specifically tailored for a given mechanism, such as DroidRA to address reflection. It also includes systematic approaches, such as EdgeMiner, capturing all callbacks in the Android framework systematically. (4) Modeling entry point methods challenges call graph construction which jeopardizes soundness.

Building on tools that have been successfully used in the study of rational billiards, such as induced maps and interval exchange transformations, we provide a construction of a one-parameter family of isosceles triangles exhibiting non-periodic trajectories that are not everywhere dense. This provides, by elementary means, a definitive negative answer to a long-standing open question on the density of non-periodic trajectories in triangular billiards., Comment: 15 pages, 2 figures

Machine learning is increasingly used for intrusion detection in IoT networks. This paper explores the effectiveness of using individual packet features (IPF), which are attributes extracted from a single network packet, such as timing, size, and source-destination information. Through literature review and experiments, we identify the limitations of IPF, showing they can produce misleadingly high detection rates. Our findings emphasize the need for approaches that consider packet interactions for robust intrusion detection. Additionally, we demonstrate that models based on IPF often fail to generalize across datasets, compromising their reliability in diverse IoT environments., Comment: 7 pages, 7 figures, 3 tables

Modern code review is a process in which an incremental code contribution made by a code author is reviewed by one or more peers before it is committed to the version control system. An important element of modern code review is verifying that code contributions adhere to best practices. While some of these best practices can be automatically verified, verifying others is commonly left to human reviewers. This paper reports on the development, deployment, and evaluation of AutoCommenter, a system backed by a large language model that automatically learns and enforces coding best practices. We implemented AutoCommenter for four programming languages (C++, Java, Python, and Go) and evaluated its performance and adoption in a large industrial setting. Our evaluation shows that an end-to-end system for learning and enforcing coding best practices is feasible and has a positive impact on the developer workflow. Additionally, this paper reports on the challenges associated with deploying such a system to tens of thousands of developers and the corresponding lessons learned., Comment: To appear at the ACM International Conference on AI-Powered Software (AIware '24)

This study investigates human and ChatGPT text simplification and its relationship to dependency distance. A set of 220 sentences, with increasing grammatical difficulty as measured in a prior user study, were simplified by a human expert and using ChatGPT. We found that the three sentence sets all differed in mean dependency distances: the highest in the original sentence set, followed by ChatGPT simplified sentences, and the human simplified sentences showed the lowest mean dependency distance.

Digital technologies have positively transformed society, but they have also led to undesirable consequences not anticipated at the time of design or development. We posit that insights into past undesirable consequences can help researchers and practitioners gain awareness and anticipate potential adverse effects. To test this assumption, we introduce BLIP, a system that extracts real-world undesirable consequences of technology from online articles, summarizes and categorizes them, and presents them in an interactive, web-based interface. In two user studies with 15 researchers in various computer science disciplines, we found that BLIP substantially increased the number and diversity of undesirable consequences they could list in comparison to relying on prior knowledge or searching online. Moreover, BLIP helped them identify undesirable consequences relevant to their ongoing projects, made them aware of undesirable consequences they "had never considered," and inspired them to reflect on their own experiences with technology., Comment: To appear in the Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI '24), May 11--16, 2024, Honolulu, HI, USA

This work focuses on leveraging and selecting from vast, unlabeled, open data to pre-fine-tune a pre-trained language model. The goal is to minimize the need for costly domain-specific data for subsequent fine-tuning while achieving desired performance levels. While many data selection algorithms have been designed for small-scale applications, rendering them unsuitable for our context, some emerging methods do cater to language data scales. However, they often prioritize data that aligns with the target distribution. While this strategy may be effective when training a model from scratch, it can yield limited results when the model has already been pre-trained on a different distribution. Differing from prior work, our key idea is to select data that nudges the pre-training distribution closer to the target distribution. We show the optimality of this approach for fine-tuning tasks under certain conditions. We demonstrate the efficacy of our methodology across a diverse array of tasks (NLU, NLG, zero-shot) with models up to 2.7B, showing that it consistently surpasses other selection methods. Moreover, our proposed method is significantly faster than existing techniques, scaling to millions of samples within a single GPU hour. Our code is open-sourced (Code repository: https://anonymous.4open.science/r/DV4LLM-D761/ ). While fine-tuning offers significant potential for enhancing performance across diverse tasks, its associated costs often limit its widespread adoption; with this work, we hope to lay the groundwork for cost-effective fine-tuning, making its benefits more accessible., Comment: Published as a conference paper at ICLR 2024

In a previous paper using Gaia DR2 data, we demonstrated that the two closely situated open clusters Collinder 135 and UBC 7 might have formed together about 50 Myr ago. In this work, we performed star-by-star dynamical modelling of the evolution of the open clusters Collinder 135 and UBC 7 from their supposed initial state to their present-day state, reproducing observational distributions of members. Modelling of the Collinder 135 and UBC 7 dynamical evolution was done using the high-order parallel N-body code \phi-GPU with up-to-date stellar evolution. Membership and characteristics of the clusters were acquired based on Gaia DR3 data. The comparison of the present-day radial cumulative star count obtained from the N-body simulations with the current observational data gave us full consistency of the model with observational data, especially in the central 8 pc, where 80% of the stars reside. The proper motion velocity components obtained from the N-body simulations of the stars are also quite consistent with the observed distributions and error bars. These results show that our numerical modelling is able to reproduce the open clusters' current complex 6D observed phase-space distributions with a high level of confidence. Thus, the model demonstrates that the hypothesis of a common origin of Collinder 135 and UBC 7 complies with present-day observational data., Comment: Astronomy & Astrophysics, accepted

Extended dynamic mode decomposition (EDMD) is a data-driven algorithm for approximating spectral data of the Koopman operator associated to a dynamical system, combining a Galerkin method of order N and collocation method of order M. Spectral convergence of this method subtly depends on appropriate choice of the space of observables. For chaotic analytic full branch maps of the interval, we derive a constraint between M and N guaranteeing spectral convergence of EDMD., Comment: 24 pages

Quantum computing is an exciting field with high disruptive potential, but very difficult to access. For this reason, many approaches to teaching quantum computing are being developed worldwide. This always raises questions about the didactic concept, the content actually taught, and how to measure the success of the teaching concept. In 2022 and 2023, the authors taught a total of nine two-week MOOCs (massive open online courses) with different possible learning paths on the Hasso Plattner Institute's OpenHPI platform. The purpose of the platform is to make computer science education available to everyone free of charge. The nine quantum courses form a self-contained curriculum. A total of more than 17{,}000 course attendances have been taken by about 7400 natural persons, and the number is still rising. This paper presents the course concept and evaluates the anonymized data on the background of the participants, their behaviour in the courses, and their learning success. This paper is the first to analyze such a large dataset of MOOC-based quantum computing education. The summarized results are a heterogeneous personal background of the participants biased towards IT professionals, a majority following the didactic recommendations, and a high success rate, which is strongly correlatated with following the didactic recommendations. The amount of data from such a large group of quantum computing learners provides many avenues for further research in the field of quantum computing education. The analyses show that the MOOCs are a low-threshold concept for getting into quantum computing. It was very well received by the participants. The concept can serve as an entry point and guide for the design of quantum computing courses.

The rapid proliferation of digital content and the ever-growing need for precise object recognition and segmentation have driven the advancement of cutting-edge techniques in the field of object classification and segmentation. This paper introduces "Learn and Search", a novel approach for object lookup that leverages the power of contrastive learning to enhance the efficiency and effectiveness of retrieval systems. In this study, we present an elegant and innovative methodology that integrates deep learning principles and contrastive learning to tackle the challenges of object search. Our extensive experimentation reveals compelling results, with "Learn and Search" achieving superior Similarity Grid Accuracy, showcasing its efficacy in discerning regions of utmost similarity within an image relative to a cropped image. The seamless fusion of deep learning and contrastive learning to address the intricacies of object identification not only promises transformative applications in image recognition, recommendation systems, and content tagging but also revolutionizes content-based search and retrieval. The amalgamation of these techniques, as exemplified by "Learn and Search," represents a significant stride in the ongoing evolution of methodologies in the dynamic realm of object classification and segmentation., Comment: 9 pages, 4 figures

Chaotic hyperbolic dynamical systems enjoy a surprising degree of rigidity, a fact which is well known in the mathematics community but perhaps less so in theoretical physics circles. Low-dimensional hyperbolic systems are either conjugate to linear automorphisms, that is, dynamically equivalent to the Arnold cat map and its variants, or their hyperbolic structure is not smooth. We illustrate this dichotomy using a family of analytic maps, for which we show by means of numerical simulations that the corresponding hyperbolic structure is not smooth, thereby providing an example for a global mechanism which produces non-smooth phase space structures in an otherwise smooth dynamical system., Comment: To appear in Nonlinearity

The Jiao Tong University Spectroscopic Telescope (JUST) is a 4.4-meter f/6.0 segmentedmirror telescope dedicated to spectroscopic observations. The JUST primary mirror is composed of 18 hexagonal segments, each with a diameter of 1.1 m. JUST provides two Nasmyth platforms for placing science instruments. One Nasmyth focus fits a field of view of 10 arcmin and the other has an extended field of view of 1.2 deg with correction optics. A tertiary mirror is used to switch between the two Nasmyth foci. JUST will be installed at a site at Lenghu in Qinghai Province, China, and will conduct spectroscopic observations with three types of instruments to explore the dark universe, trace the dynamic universe, and search for exoplanets: (1) a multi-fiber (2000 fibers) medium-resolution spectrometer (R=4000-5000) to spectroscopically map galaxies and large-scale structure; (2) an integral field unit (IFU) array of 500 optical fibers and/or a long-slit spectrograph dedicated to fast follow-ups of transient sources for multimessenger astronomy; (3) a high-resolution spectrometer (R~100000) designed to identify Jupiter analogs and Earth-like planets, with the capability to characterize the atmospheres of hot exoplanets., Comment: 28 pages, 6 figures

Training image-based object detectors presents formidable challenges, as it entails not only the complexities of object detection but also the added intricacies of precisely localizing objects within potentially diverse and noisy environments. However, the collection of imagery itself can often be straightforward; for instance, cameras mounted in vehicles can effortlessly capture vast amounts of data in various real-world scenarios. In light of this, we introduce a groundbreaking method for training single-stage object detectors through unsupervised/self-supervised learning. Our state-of-the-art approach has the potential to revolutionize the labeling process, substantially reducing the time and cost associated with manual annotation. Furthermore, it paves the way for previously unattainable research opportunities, particularly for large, diverse, and challenging datasets lacking extensive labels. In contrast to prevalent unsupervised learning methods that primarily target classification tasks, our approach takes on the unique challenge of object detection. We pioneer the concept of intra-image contrastive learning alongside inter-image counterparts, enabling the acquisition of crucial location information essential for object detection. The method adeptly learns and represents this location information, yielding informative heatmaps. Our results showcase an outstanding accuracy of \textbf{89.2\%}, marking a significant breakthrough of approximately \textbf{15x} over random initialization in the realm of unsupervised object detection within the field of computer vision., Comment: 10 pages, 5 figures

FPGA technology mapping is the process of implementing a hardware design expressed in high-level HDL (hardware design language) code using the low-level, architecture-specific primitives of the target FPGA. As FPGAs become increasingly heterogeneous, achieving high performance requires hardware synthesis tools that better support mapping to complex, highly configurable primitives like digital signal processors (DSPs). Current tools support DSP mapping via handwritten special-case mapping rules, which are laborious to write, error-prone, and often overlook mapping opportunities. We introduce Lakeroad, a principled approach to technology mapping via sketch-guided program synthesis. Lakeroad leverages two techniques -- architecture-independent sketch templates and semantics extraction from HDL -- to provide extensible technology mapping with stronger correctness guarantees and higher coverage of mapping opportunities than state-of-the-art tools. Across representative microbenchmarks, Lakeroad produces 2--3.5$\times$ the number of optimal mappings compared to proprietary state-of-the-art tools and 6--44$\times$ the number of optimal mappings compared to popular open-source tools, while also providing correctness guarantees not given by any other tool.

Based on the astrometry and photometry in Gaia DR3, we identified new nearby white dwarfs and validated those that had been missed from recent white dwarf catalogues despite being previously documented. To ensure the reliability of their astrometric solutions, we used a cut on just two parameters from Gaia DR3: the amplitude of the image parameter determination goodness-of-fit and the parallax-over-error ratio. In addition, we imposed photometric signal-to-noise requirements to ensure the reliable identification of white dwarfs when using the colour-magnitude diagram. We have identified nine previously unreported white dwarfs within the local population of 50 pc, and validated 21 previously reported white dwarfs missing from the GCWD21 (Gentile Fusillo et al. 2021) and other recent volume-limited white dwarf samples. A few of these objects belong to the rare class of ultra-cool white dwarfs. Four white dwarfs in our sample have an effective temperature of $T_{eff}\leq4000$ K within the $1\sigma$ interval, and two of them have an absolute magnitude of $M_G > 16.0$ mag. The identified white dwarfs are predominantly located in crowded fields, such as near the Galactic plane or in the foreground of the Large Magellanic Cloud. We also find that 19 of these white dwarfs have common proper motion companions with angular separations ranging from $1.1''$ to $7.1''$ and brightness differences between the components of up to 9.8 magnitudes. One of these systems is a triple system consisting of a white dwarf and two K dwarfs, while another is a double white dwarf system. We have identified 103 contaminants among the 2338 high-confidence white dwarfs in the 50 pc subsample of the GCWD21 and have found that their astrometric solutions in Gaia DR3 are spurious, improving the purity by 4.4%., Comment: Accepted for publication in A&A on 02 December 2023. 15 pages, 14 figures. Abstract abridged due to arXiv's 1920 character limit

Statistical models should accurately reflect analysts' domain knowledge about variables and their relationships. While recent tools let analysts express these assumptions and use them to produce a resulting statistical model, it remains unclear what analysts want to express and how externalization impacts statistical model quality. This paper addresses these gaps. We first conduct an exploratory study of analysts using a domain-specific language (DSL) to express conceptual models. We observe a preference for detailing how variables relate and a desire to allow, and then later resolve, ambiguity in their conceptual models. We leverage these findings to develop rTisane, a DSL for expressing conceptual models augmented with an interactive disambiguation process. In a controlled evaluation, we find that rTisane's DSL helps analysts engage more deeply with and accurately externalize their assumptions. rTisane also leads to statistical models that match analysts' assumptions, maintain analysis intent, and better fit the data.

Previous research on behaviour-based attack detection on networks of IoT devices has resulted in machine learning models whose ability to adapt to unseen data is limited, and often not demonstrated. In this paper we present an approach for modelling IoT network attacks that focuses on generalizability, yet also leads to better detection and performance. First, we present an improved rolling window approach for feature extraction, and introduce a multi-step feature selection process that reduces overfitting. Second, we build and test models using isolated train and test datasets, thereby avoiding common data leaks that have limited the generalizability of previous models. Third, we rigorously evaluate our methodology using a diverse portfolio of machine learning models, evaluation metrics and datasets. Finally, we build confidence in the models by using explainable AI techniques, allowing us to identify the features that underlie accurate detection of attacks., Comment: 25 pages (13 main, 12 supplementary appendix), 20 figures, 14 tables

Recent analysis of the kilonova, AT2017gfo, has indicated that this event was highly spherical. This may challenge hydrodynamics simulations of binary neutron star mergers, which usually predict a range of asymmetries, and radiative transfer simulations show a strong direction dependence. Here we investigate whether the synthetic spectra from a 3D kilonova simulation of asymmetric ejecta from a hydrodynamical merger simulation can be compatible with the observational constraints suggesting a high degree of sphericity in AT2017gfo. Specifically, we determine whether fitting a simple P-Cygni line profile model leads to a value for the photospheric velocity that is consistent with the value obtained from the expanding photosphere method. We would infer that our kilonova simulation is highly spherical at early times, when the spectra resemble a blackbody distribution. The two independently inferred photospheric velocities can be very similar, implying a high degree of sphericity, which can be as spherical as inferred for AT2017gfo, demonstrating that the photosphere can appear spherical even for asymmetrical ejecta. The last-interaction velocities of radiation escaping the simulation show a high degree of sphericity, supporting the inferred symmetry of the photosphere. We find that when the synthetic spectra resemble a blackbody the expanding photosphere method can be used to obtain an accurate luminosity distance (within 4-7 per cent)., Comment: Accepted by MNRAS

Airborne LiDAR systems have the capability to capture the Earth's surface by generating extensive point cloud data comprised of points mainly defined by 3D coordinates. However, labeling such points for supervised learning tasks is time-consuming. As a result, there is a need to investigate techniques that can learn from unlabeled data to significantly reduce the number of annotated samples. In this work, we propose to train a self-supervised encoder with Barlow Twins and use it as a pre-trained network in the task of semantic scene segmentation. The experimental results demonstrate that our unsupervised pre-training boosts performance once fine-tuned on the supervised task, especially for under-represented categories., Comment: International conference Machine Vision Applications 2023

For a pair of supermassive black holes (SMBHs) in the remnant of a dual galaxy merger, well-known models exist to describe their dynamical evolution until the final coalescence accompanied by the emission of a low-frequency gravitational wave (GW) signal. In this article, we investigate the dynamical evolution of three SMBH triple systems recovered from the ROMULUS25 cosmological simulation to explore common dynamical evolution patterns and assess typical coalescence times. For this purpose, we construct initial conditions from the ROMULUS25 data and perform high-resolution gravitodynamical \N-body simulations. We track the orbital evolution from the galactic inspiral to the formation of hard binaries at sub-parsec separation and use the observed hardening rates to project the time of coalescence. In all cases, the two heaviest black holes form an efficiently hardening binary that merges within fractions of the Hubble time. The lightest SMBH either gets ejected, forms a stable hierarchical triple system with the heavier binary, forms a hardening binary with the previously merged binary's remnant, or remains on a wide galactic orbit. The coalescence times of the lighter black holes are thus significantly longer than for the heavier binary, as they experience lower dynamical friction and stellar hardening rates. We observe the formation of hierarchical triples when the density profile of the galactic nucleus is sufficiently steep., Comment: 11 pages, 6 figures, 4 tables, accepted for publication in A&A

We show that spectral data of the Koopman operator arising from an analytic expanding circle map $\tau$ can be effectively calculated using an EDMD-type algorithm combining a collocation method of order m with a Galerkin method of order n. The main result is that if $m \geq \delta n$, where $\delta$ is an explicitly given positive number quantifying by how much $\tau$ expands concentric annuli containing the unit circle, then the method converges and approximates the spectrum of the Koopman operator, taken to be acting on a space of analytic hyperfunctions, exponentially fast in n. Additionally, these results extend to more general expansive maps on suitable annuli containing the unit circle., Comment: 18 pages, 4 figures