Search

Modern artificial intelligence (AI) systems are powered by foundation models. This paper presents a new set of foundation models, called Llama 3. It is a herd of language models that natively support multilinguality, coding, reasoning, and tool usage. Our largest model is a dense Transformer with 405B parameters and a context window of up to 128K tokens. This paper presents an extensive empirical evaluation of Llama 3. We find that Llama 3 delivers comparable quality to leading language models such as GPT-4 on a plethora of tasks. We publicly release Llama 3, including pre-trained and post-trained versions of the 405B parameter language model and our Llama Guard 3 model for input and output safety. The paper also presents the results of experiments in which we integrate image, video, and speech capabilities into Llama 3 via a compositional approach. We observe this approach performs competitively with the state-of-the-art on image, video, and speech recognition tasks. The resulting models are not yet being broadly released as they are still under development.

Model reduction is an active research field to construct low-dimensional surrogate models of high fidelity to accelerate engineering design cycles. In this work, we investigate model reduction for linear structured systems using dominant reachable and observable subspaces. When the training set $-$ containing all possible interpolation points $-$ is large, then these subspaces can be determined by solving many large-scale linear systems. However, for high-fidelity models, this easily becomes computationally intractable. To circumvent this issue, in this work, we propose an active sampling strategy to sample only a few points from the given training set, which can allow us to estimate those subspaces accurately. To this end, we formulate the identification of the subspaces as the solution of the generalized Sylvester equations, guiding us to select the most relevant samples from the training set to achieve our goals. Consequently, we construct solutions of the matrix equations in low-rank forms, which encode subspace information. We extensively discuss computational aspects and efficient usage of the low-rank factors in the process of obtaining reduced-order models. We illustrate the proposed active sampling scheme to obtain reduced-order models via dominant reachable and observable subspaces and present its comparison with the method where all the points from the training set are taken into account. It is shown that the active sample strategy can provide us $17$x speed-up without sacrificing any noticeable accuracy., Comment: 20 pages, 9 figures

Autonomous agent applications powered by large language models (LLMs) have recently risen to prominence as effective tools for addressing complex real-world tasks. At their core, agentic workflows rely on LLMs to plan and execute the use of tools and external Application Programming Interfaces (APIs) in sequence to arrive at the answer to a user's request. Various benchmarks and leaderboards have emerged to evaluate an LLM's capabilities for tool and API use; however, most of these evaluations only track single or multiple isolated API calling capabilities. In this paper, we present NESTFUL, a benchmark to evaluate LLMs on nested sequences of API calls, i.e., sequences where the output of one API call is passed as input to a subsequent call. NESTFUL has a total of 300 human annotated samples divided into two types - executable and non-executable. The executable samples are curated manually by crawling Rapid-APIs whereas the non-executable samples are hand picked by human annotators from data synthetically generated using an LLM. We evaluate state-of-the-art LLMs with function calling abilities on NESTFUL. Our results show that most models do not perform well on nested APIs in NESTFUL as compared to their performance on the simpler problem settings available in existing benchmarks.

The purpose of the current work is the development and comparison of Fourier neural operators (FNOs) for surrogate modeling of the quasi-static mechanical response of polycrystalline materials. Three types of such FNOs are considered here: a physics-guided FNO (PgFNO), a physics-informed FNO (PiFNO), and a physics-encoded FNO (PeFNO). These are trained and compared with the help of stress field data from a reference model for heterogeneous elastic materials with a periodic grain microstructure. Whereas PgFNO training is based solely on these data, that of the PiFNO and PeFNO is in addition constrained by the requirement that stress fields satisfy mechanical equilibrium, i.e., be divergence-free. The difference between the PiFNO and PeFNO lies in how this constraint is taken into account; in the PiFNO, it is included in the loss function, whereas in the PeFNO, it is "encoded" in the operator architecture. In the current work, this encoding is based on a stress potential and Fourier transforms. As a result, only the training of the PiFNO is constrained by mechanical equilibrium; in contrast, mechanical equilibrium constrains both the training and output of the PeFNO. Due in particular to this, stress fields calculated by the trained PeFNO are significantly more accurate than those calculated by the trained PiFNO in the example cases considered., Comment: 17 pages, 11 figures

We introduce Instruct-SkillMix, an automated approach for creating diverse, high quality SFT data. The Instruct-SkillMix pipeline involves two stages, each leveraging an existing powerful LLM: (1) Skill extraction: uses the LLM to extract core "skills" for instruction-following, either from existing datasets, or by directly prompting the model; (2) Data generation: uses the powerful LLM to generate (instruction, response) data that exhibit a randomly chosen pair of these skills. Here, the use of random skill combinations promotes diversity and difficulty. Vanilla SFT (i.e., no PPO, DPO, or RL methods) on data generated from Instruct-SkillMix leads to strong gains on instruction following benchmarks such as AlpacaEval 2.0, MT-Bench, and WildBench. With just $4$K examples, LLaMA-3-8B-Base achieves 42.76% length-controlled win rate on AlpacaEval 2.0. To our knowledge, this achieves state-of-the-art performance among all models that have only undergone SFT (no RL methods) and competes with proprietary models such as Claude 3 Opus and LLaMA-3.1-405B-Instruct. Ablation studies also suggest plausible reasons for why creating open instruction-tuning datasets via naive crowd-sourcing has proved difficult. Introducing low quality answers ("shirkers") in $20\%$ of Instruct-SkillMix examples causes performance to plummet, sometimes catastrophically. The Instruct-SkillMix pipeline is flexible and is adaptable to other settings.

While Endpoint Detection and Response (EDR) are able to efficiently monitor threats by comparing static rules to the event stream, their inability to incorporate past system context leads to high rates of false alarms. Recent work has demonstrated Provenance-based Intrusion Detection Systems (Prov-IDS) that can examine the causal relationships between abnormal behaviors to improve threat classification. However, employing these Prov-IDS in practical settings remains difficult -- state-of-the-art neural network based systems are only fast in a fully offline deployment model that increases attacker dwell time, while simultaneously using simplified and less accurate provenance graphs to reduce memory consumption. Thus, today's Prov-IDS cannot operate effectively in the real-time streaming setting required for commercial EDR viability. This work presents the design and implementation of ORCHID, a novel Prov-IDS that performs fine-grained detection of process-level threats over a real time event stream. ORCHID takes advantage of the unique immutable properties of a versioned provenance graphs to iteratively embed the entire graph in a sequential RNN model while only consuming a fraction of the computation and memory costs. We evaluate ORCHID on four public datasets, including DARPA TC, to show that ORCHID can provide competitive classification performance while eliminating detection lag and reducing memory consumption by two orders of magnitude.

Deliberation involves participants exchanging knowledge, arguments, and perspectives and has been shown to be effective at addressing polarization. The Stanford Online Deliberation Platform facilitates large-scale deliberations. It enables video-based online discussions on a structured agenda for small groups without requiring human moderators. This paper's data comes from various deliberation events, including one conducted in collaboration with Meta in 32 countries, and another with 38 post-secondary institutions in the US. Estimating the quality of contributions in a conversation is crucial for assessing feature and intervention impacts. Traditionally, this is done by human annotators, which is time-consuming and costly. We use a large language model (LLM) alongside eight human annotators to rate contributions based on justification, novelty, expansion of the conversation, and potential for further expansion, with scores ranging from 1 to 5. Annotators also provide brief justifications for their ratings. Using the average rating from other human annotators as the ground truth, we find the model outperforms individual human annotators. While pairs of human annotators outperform the model in rating justification and groups of three outperform it on all four metrics, the model remains competitive. We illustrate the usefulness of the automated quality rating by assessing the effect of nudges on the quality of deliberation. We first observe that individual nudges after prolonged inactivity are highly effective, increasing the likelihood of the individual requesting to speak in the next 30 seconds by 65%. Using our automated quality estimation, we show that the quality ratings for statements prompted by nudging are similar to those made without nudging, signifying that nudging leads to more ideas being generated in the conversation without losing overall quality.

We collected more than 300 high-resolution spectra of the 300 Myr old star BD+40 2790 (TOI-2076) over ~3 years. This star hosts three transiting planets discovered by TESS, with orbital periods ~10, 21, and 35 days. BD+40 2790 shows an activity-induced scatter larger than 30 m/s in the radial velocities. We employed different methods to measure the stellar radial velocities and several models to filter out the dominant stellar activity signal, in order to bring to light the planet-induced signals which are expected to have semi-amplitudes one order of magnitude lower. We evaluated the mass loss rate of the planetary atmospheres using photoionization hydrodynamic modeling. The dynamical analysis confirms that the three sub-Neptune-sized companions (our radius measurements are $R_b$=2.54$\pm$0.04, $R_c$=3.35$\pm$0.05, and $R_d$=3.29$\pm$0.06 $R_{\rm Earth}$) have masses in the planetary regime. We derive 3$\sigma$ upper limits below or close to the mass of Neptune for all the planets: 11--12, 12--13.5, and 14--19 $M_{\rm Earth}$ for planet $b$, $c$, and $d$ respectively. In the case of planet $d$, we found promising clues that the mass could be between ~7 and 8 $M_{\rm Earth}$, with a significance level between 2.3--2.5$\sigma$ (at best). This result must be further investigated using other analysis methods or using high-precision near-IR spectrographs to collect new radial velocities, which could be less affected by stellar activity. Atmospheric photo-evaporation simulations predict that BD+40~2790 b is currently losing its H-He gaseous envelope, which will be completely lost at an age within 0.5--3 Gyr if its current mass is lower than 12 $M_{\rm Earth}$. BD+40 2790 c could have a lower bulk density than $b$, and it could retain its atmosphere up to an age of 5 Gyr. For the outermost planet $d$, we predict almost negligible evolution of its mass and radius induced by photo-evaporation., Comment: Accepted for publication on A&A. Abstract abridged

In the context of increasingly complex environmental challenges, effective pollution control mechanisms are crucial. By extending the state of the art auction mechanisms, we aim to develop an efficient approach for allocating pollution abatement resources in a multi-pollutant setting with pollutants affecting each other's reduction costs. We modify the Combinatorial Multi-Round Ascending Auction for the auction of escape permits of pollutants with co-dependent reduction processes, specifically, greenhouse gas emissions and nutrient runoff in Finnish agriculture. We show the significant advantages of this mechanism in pollution control through experiments on the bid prices and amount of escape permits sold in multiple auction simulations.

This survey examines the broad suite of methods and models for combining machine learning with physics knowledge for prediction and forecast, with a focus on partial differential equations. These methods have attracted significant interest due to their potential impact on advancing scientific research and industrial practices by improving predictive models with small- or large-scale datasets and expressive predictive models with useful inductive biases. The survey has two parts. The first considers incorporating physics knowledge on an architectural level through objective functions, structured predictive models, and data augmentation. The second considers data as physics knowledge, which motivates looking at multi-task, meta, and contextual learning as an alternative approach to incorporating physics knowledge in a data-driven fashion. Finally, we also provide an industrial perspective on the application of these methods and a survey of the open-source ecosystem for physics-informed machine learning., Comment: 56 pages, 8 figures, 2 tables

In modern deep learning, the models are learned by applying gradient updates using an optimizer, which transforms the updates based on various statistics. Optimizers are often hand-designed and tuning their hyperparameters is a big part of the training process. Learned optimizers have shown some initial promise, but are generally unsuccessful as a general optimization mechanism applicable to every problem. In this work we explore a different direction: instead of learning general optimizers, we instead specialize them to a specific training environment. We propose a novel optimizer technique that learns a layer-specific linear combination of update directions provided by a set of base optimizers, effectively adapting its strategy to the specific model and dataset. When evaluated on image classification tasks, this specialized optimizer significantly outperforms both traditional off-the-shelf methods such as Adam, as well as existing general learned optimizers. Moreover, it demonstrates robust generalization with respect to model initialization, evaluating on unseen datasets, and training durations beyond its meta-training horizon.

The goal of object-centric representation learning is to decompose visual scenes into a structured representation that isolates the entities. Recent successes have shown that object-centric representation learning can be scaled to real-world scenes by utilizing pre-trained self-supervised features. However, so far, object-centric methods have mostly been applied in-distribution, with models trained and evaluated on the same dataset. This is in contrast to the wider trend in machine learning towards general-purpose models directly applicable to unseen data and tasks. Thus, in this work, we study current object-centric methods through the lens of zero-shot generalization by introducing a benchmark comprising eight different synthetic and real-world datasets. We analyze the factors influencing zero-shot performance and find that training on diverse real-world images improves transferability to unseen scenarios. Furthermore, inspired by the success of task-specific fine-tuning in foundation models, we introduce a novel fine-tuning strategy to adapt pre-trained vision encoders for the task of object discovery. We find that the proposed approach results in state-of-the-art performance for unsupervised object discovery, exhibiting strong zero-shot transfer to unseen datasets.

Lead chalcogenides are the promising thermoelectric (TE) materials having narrow band gap. The present work investigates the TE behaviour of PbSe in the temperature range 300-500 K. The transport properties of the sample have been studied using the Abinit and BoltzTrap code. The experimentally observed value of \textit{S} at 300 and 500 K is found to be $\sim$ 198 and 266 $\mu$V K$^{-1}$, respectively. The rate of increase in \emph{S} from 300 to 460 (460 to 500) K is found to be $\sim$ 0.4 (0.09). The temperature dependent electrical conductivity \textit{($\sigma$)} shows the increasing trend, with values of $\sim $ 0.35 $\times $ 10$^{3}$ and $\sim$ 0.58 $\times$ 10$^{3}$ $\Omega$$^{-1}$ m$^{-1}$ at 300 and 500 K, respectively. Further, the value of thermal conductivity \textit{($\kappa$)} at 300 (500) K is found to be 0.74 (1.07) W m$^{-1}$ K$^{-1}$. The value of \textit{$\kappa$} is found to be increasing upto 460 K and then starts decreasing. The dispersion plot indicates that PbSe is a direct band gap semiconductor with band gap value of 0.16 (0.27) eV considering spin-orbit coupling (without SOC). The partial density of states (PDOS) plot shows that Pb 6p and Se 4p states have a major contribution in the transport properties. The observed and calculated values of \textit{S} gives a good match for SOC case. The calculated \textit{$\sigma$} and electronic part of thermal conductivity (\textit{$\kappa{_e}$}) gives good match with the experimental data. The maximum power factor (PF) value of $\sim$ 4.3 $\times$ 10$^{-5}$ W/mK$^{2}$ is observed at 500 K. This work helps in understanding the TE behaviour of PbSe through a novel and insightful alliance of experimental measurements and DFT approach.

The ever-increasing volume of digital information necessitates efficient methods for users to extract key insights from lengthy documents. Aspect-based summarization offers a targeted approach, generating summaries focused on specific aspects within a document. Despite advancements in aspect-based summarization research, there is a continuous quest for improved model performance. Given that large language models (LLMs) have demonstrated the potential to revolutionize diverse tasks within natural language processing, particularly in the problem of summarization, this paper explores the potential of fine-tuning LLMs for the aspect-based summarization task. We evaluate the impact of fine-tuning open-source foundation LLMs, including Llama2, Mistral, Gemma and Aya, on a publicly available domain-specific aspect based summary dataset. We hypothesize that this approach will enable these models to effectively identify and extract aspect-related information, leading to superior quality aspect-based summaries compared to the state-of-the-art. We establish a comprehensive evaluation framework to compare the performance of fine-tuned LLMs against competing aspect-based summarization methods and vanilla counterparts of the fine-tuned LLMs. Our work contributes to the field of aspect-based summarization by demonstrating the efficacy of fine-tuning LLMs for generating high-quality aspect-based summaries. Furthermore, it opens doors for further exploration of using LLMs for targeted information extraction tasks across various NLP domains.

Current LLM training positions mathematical reasoning as a core capability. With publicly available sources fully tapped, there is unmet demand for diverse and challenging math questions. Relying solely on human experts is both time-consuming and costly, while LLM-generated questions often lack the requisite diversity and difficulty. We present a design framework that combines the strengths of LLMs with a human-in-the-loop approach to generate a diverse array of challenging math questions. We leverage LLM metacognition skills [Didolkar et al., 2024] of a strong LLM to extract core "skills" from existing math datasets. These skills serve as the basis for generating novel and difficult questions by prompting the LLM with random pairs of core skills. The use of two different skills within each question makes finding such questions an "out of distribution" task for both LLMs and humans. Our pipeline employs LLMs to iteratively generate and refine questions and solutions through multiturn prompting. Human annotators then verify and further refine the questions, with their efficiency enhanced via further LLM interactions. Applying this pipeline on skills extracted from the MATH dataset [Hendrycks et al., 2021] resulted in MATH$^2$ - a dataset of higher-quality math questions, as evidenced by: (a) Lower performance of all models on MATH$^2$ than on MATH (b) Higher performance on MATH when using MATH$^2$ questions as in-context examples. Although focused on mathematics, our methodology seems applicable to other domains requiring structured reasoning, and potentially as a component of scalable oversight. Also of interest is a striking relationship observed between models' performance on the new dataset: the success rate on MATH$^2$ is the square on MATH, suggesting that successfully solving the question in MATH$^2$ requires a nontrivial combination of two distinct math skills.

While hallucinations of large language models (LLMs) prevail as a major challenge, existing evaluation benchmarks on factuality do not cover the diverse domains of knowledge that the real-world users of LLMs seek information about. To bridge this gap, we introduce WildHallucinations, a benchmark that evaluates factuality. It does so by prompting LLMs to generate information about entities mined from user-chatbot conversations in the wild. These generations are then automatically fact-checked against a systematically curated knowledge source collected from web search. Notably, half of these real-world entities do not have associated Wikipedia pages. We evaluate 118,785 generations from 15 LLMs on 7,919 entities. We find that LLMs consistently hallucinate more on entities without Wikipedia pages and exhibit varying hallucination rates across different domains. Finally, given the same base models, adding a retrieval component only slightly reduces hallucinations but does not eliminate hallucinations.

Molecular HIV Surveillance (MHS) has been described as key to enabling rapid responses to HIV outbreaks. It operates by linking individuals with genetically similar viral sequences, which forms a network. A major limitation of MHS is that it depends on sequence collection, which very rarely covers the entire population of interest. Ignoring missing data by conducting complete case analysis--which assumes that the observed network is complete--has been shown to result in significantly biased estimates of network properties. We use MHS to investigate disease dynamics of the HIV epidemic in Miami-Dade County (MDC) among men who have sex with men (MSM)--only 30.1% have a reported sequence. To do so, we present an approach for making Bayesian inferences on partially observed networks. Through a simulation study, we demonstrate a reduction in error of 43%-63% between our estimates and complete case analyses. We estimate increased mixing between MSM communities in MDC, defined by race and transmission risk compared to the results based on complete case analysis. Our approach makes use of a flexible network model--congruence class model--to overcome the high computational burden of previously reported Bayesian approaches to estimate network properties from partially observed networks., Comment: 19 pages; 6 figures; 2 tables

The V1298 Tau system (20-30Myr), is a benchmark young multi-planet system that provides the opportunity to perform comparative exoplanetology between planets orbiting the same star right after their formation. We present the first atmospheric comparison between two planets in the same transiting system: V1298 Tau b and V1298 Tau c. We derive constraints on the mass of planet b and c (<20M$_\oplus$ at 3$\sigma$ confidence level and $17_{-6}^{+13} M_{\oplus}$ respectively) and atmospheric metallicity (logZ/Z$_\odot$=-2.04$_{-0.59}^{0.69}$, -0.16$_{-0.94}^{1.15}$ respectively) from atmospheric retrievals. The V1298 Tau planets, are likely to be similar in terms of mass at the current age, implying that both planets are potential sub-Neptune/super-Earth progenitors. However, planet c is expected to lose a higher fraction of its mass compared to planet b given its close proximity to the host star. Alternatively, the observed spectrum of planet c can be explained by atmospheric hazes, which is in contrast to planet b where efficient haze formation can be ruled out. Higher haze formation efficiency in planet c could be due to differences in atmospheric composition, temperature and higher UV flux incident compared to planet b., Comment: Submitted to Astronomy ad Astrophysics

In some fields, scientific data formats differ across experiments due to specialized hardware and data acquisition systems. Researchers need to develop, document, and maintain experiment-specific analysis software to interact with these data formats. These software are often tightly coupled with a particular data format. This proliferation of custom data formats has been a prominent challenge for small to mid-scale experiments. The widespread adoption of ROOT has largely mitigated this problem for the Large Hadron Collider experiments. However, many smaller experiments continue to use custom data formats to meet specific research needs. Therefore, simplifying the process of accessing a unique data format for analysis holds immense value for scientific communities within HEP. We have added Awkward Arrays as a target language for Kaitai Struct for this purpose. Researchers can describe their custom data format in the Kaitai Struct YAML (KSY) language. The Kaitai Struct Compiler generates C++ code to fill the LayoutBuilder buffers using the KSY format. In a few steps, the Kaitai Struct Awkward Runtime API can convert the generated C++ code into a compiled Python module. Finally, the raw data can be passed to the module to produce Awkward Arrays. This paper introduces the Awkward Target for the Kaitai Struct Compiler and the Kaitai Struct Awkward Runtime API. It also demonstrates the conversion of a given KSY for a specific custom file format to Awkward Arrays., Comment: presented at ACAT2024

Assessing the quality of Natural Language Generation (NLG) outputs, such as those produced by large language models (LLMs), poses significant challenges. Traditional approaches involve either resource-intensive human evaluations or automatic metrics, which often exhibit a low correlation with human judgment. In this study, we propose Review-Feedback-Reason (ReFeR), a novel evaluation framework for NLG using LLM agents. We rigorously test ReFeR using two pre-existing benchmark datasets on diverse NLG tasks. The proposed framework not only enhances the accuracy of NLG evaluation, surpassing previous benchmarks by $\sim$20\%, but also generates constructive feedback and significantly improves collective reasoning. This feedback is then leveraged for the creation of instruction-tuning datasets, which, when used to fine-tune smaller models like Mistral-7B, makes them extremely good evaluators, yielding a better correlation with human evaluations and performance nearly on par with GPT-3.5. We highlight the effectiveness of our methodology through its application on three reasoning benchmarks, where it outperforms most of the state-of-the-art methods, and also outperforms the reasoning capabilities of models like GPT-3.5 Turbo by $\sim$11.67\% and GPT-4 by $\sim$1\% on an average., Comment: Paper Under Review

We present results from a search for X-ray/gamma-ray counterparts of gravitational-wave (GW) candidates from the third observing run (O3) of the LIGO-Virgo-KAGRA (LVK) network using the Swift Burst Alert Telescope (Swift-BAT). The search includes 636 GW candidates received in low latency, 86 of which have been confirmed by the offline analysis and included in the third cumulative Gravitational-Wave Transient Catalogs (GWTC-3). Targeted searches were carried out on the entire GW sample using the maximum--likelihood NITRATES pipeline on the BAT data made available via the GUANO infrastructure. We do not detect any significant electromagnetic emission that is temporally and spatially coincident with any of the GW candidates. We report flux upper limits in the 15-350 keV band as a function of sky position for all the catalog candidates. For GW candidates where the Swift-BAT false alarm rate is less than 10$^{-3}$ Hz, we compute the GW--BAT joint false alarm rate. Finally, the derived Swift-BAT upper limits are used to infer constraints on the putative electromagnetic emission associated with binary black hole mergers., Comment: 50 pages, 10 figures, 4 tables

Literature search questions, such as "where can I find research on the evaluation of consistency in generated summaries?" pose significant challenges for modern search engines and retrieval systems. These questions often require a deep understanding of research concepts and the ability to reason over entire articles. In this work, we introduce LitSearch, a retrieval benchmark comprising 597 realistic literature search queries about recent ML and NLP papers. LitSearch is constructed using a combination of (1) questions generated by GPT-4 based on paragraphs containing inline citations from research papers and (2) questions about recently published papers, manually written by their authors. All LitSearch questions were manually examined or edited by experts to ensure high quality. We extensively benchmark state-of-the-art retrieval models and also evaluate two LLM-based reranking pipelines. We find a significant performance gap between BM25 and state-of-the-art dense retrievers, with a 24.8% difference in absolute recall@5. The LLM-based reranking strategies further improve the best-performing dense retriever by 4.4%. Additionally, commercial search engines and research tools like Google Search perform poorly on LitSearch, lagging behind the best dense retriever by 32 points. Taken together, these results show that LitSearch is an informative new testbed for retrieval systems while catering to a real-world use case., Comment: Dataset and code available at https://github.com/princeton-nlp/LitSearch

Neural Machine Translation (NMT) remains a formidable challenge, especially when dealing with low-resource languages. Pre-trained sequence-to-sequence (seq2seq) multi-lingual models, such as mBART-50, have demonstrated impressive performance in various low-resource NMT tasks. However, their pre-training has been confined to 50 languages, leaving out support for numerous low-resource languages, particularly those spoken in the Indian subcontinent. Expanding mBART-50's language support requires complex pre-training, risking performance decline due to catastrophic forgetting. Considering these expanding challenges, this paper explores a framework that leverages the benefits of a pre-trained language model along with knowledge distillation in a seq2seq architecture to facilitate translation for low-resource languages, including those not covered by mBART-50. The proposed framework employs a multilingual encoder-based seq2seq model as the foundational architecture and subsequently uses complementary knowledge distillation techniques to mitigate the impact of imbalanced training. Our framework is evaluated on three low-resource Indic languages in four Indic-to-Indic directions, yielding significant BLEU-4 and chrF improvements over baselines. Further, we conduct human evaluation to confirm effectiveness of our approach. Our code is publicly available at https://github.com/raypretam/Two-step-low-res-NMT., Comment: Published at Seventh LoResMT Workshop at ACL 2024

Legal systems worldwide are inundated with exponential growth in cases and documents. There is an imminent need to develop NLP and ML techniques for automatically processing and understanding legal documents to streamline the legal system. However, evaluating and comparing various NLP models designed specifically for the legal domain is challenging. This paper addresses this challenge by proposing IL-TUR: Benchmark for Indian Legal Text Understanding and Reasoning. IL-TUR contains monolingual (English, Hindi) and multi-lingual (9 Indian languages) domain-specific tasks that address different aspects of the legal system from the point of view of understanding and reasoning over Indian legal documents. We present baseline models (including LLM-based) for each task, outlining the gap between models and the ground truth. To foster further research in the legal domain, we create a leaderboard (available at: https://exploration-lab.github.io/IL-TUR/) where the research community can upload and compare legal text understanding systems., Comment: Accepted at ACL 2024 Main Conference; 40 Pages (9 Pages + References + Appendix)

Gravitational lensing offers a powerful tool for exploring the matter distribution in the Universe. Thanks to their low frequencies and phase coherence, gravitational waves (GWs) allow for the observation of novel wave-optics features (WOFs) in lensing, inaccessible to electromagnetic signals. Combined with the existing accurate source models, lensed GWs can be used to infer the properties of gravitational lenses. The prospect is particularly compelling for space-borne detectors, where the high signal-to-noise ratio expected from massive black hole binary mergers allows WOFs to be distinguished deep into the weak lensing regime, drastically increasing the detection probability. Here, we investigate in detail the capacity of the LISA mission to detect WOFs caused by dark matter halos, galaxies and the supermassive black holes (SMBHs) within them. We estimate the total optical depth to be $\lambda_{\rm tot} \sim 6 \times 10 ^{-3}$ for the loudest binaries of total mass $M_{\rm BBH} \sim 10^6 M_{\odot}$, with the dominant contribution coming from SMBHs. We also find that WOFs in low-mass binaries $M_{\rm BBH} \sim 10^4 M_{\odot}$ are more likely due to the central galaxies. Within our model of gravitational lenses, we predict $\mathcal{O}(0.1)-\mathcal{O}(1)$ weakly-lensed events to be detectable during the 5 years of LISA mission, depending on the source population models. We show that WOFs signatures are very sensitive to the properties of dark-matter halos with $M_{\rm vir}\in (10^6-10^8)M_\odot$: increasing the compactness parameter by $\sim 3$ in that range raises the detection rate by $\sim 26$. Additionally, we show that collective effects from the complex inner halo structure can further enhance detectability. This suggests that lensed GWs in LISA will be an excellent probe of dark-matter theories, baryonic and halo sub-structures., Comment: 26 pages, 19 figures, 3 tables

Citing pertinent literature is pivotal to writing and reviewing a scientific document. Existing techniques mainly focus on the local context or the global context for recommending citations but fail to consider the actual human citation behaviour. We propose SymTax, a three-stage recommendation architecture that considers both the local and the global context, and additionally the taxonomical representations of query-candidate tuples and the Symbiosis prevailing amongst them. SymTax learns to embed the infused taxonomies in the hyperbolic space and uses hyperbolic separation as a latent feature to compute query-candidate similarity. We build a novel and large dataset ArSyTa containing 8.27 million citation contexts and describe the creation process in detail. We conduct extensive experiments and ablation studies to demonstrate the effectiveness and design choice of each module in our framework. Also, combinatorial analysis from our experiments shed light on the choice of language models (LMs) and fusion embedding, and the inclusion of section heading as a signal. Our proposed module that captures the symbiotic relationship solely leads to performance gains of 26.66% and 39.25% in Recall@5 w.r.t. SOTA on ACL-200 and RefSeer datasets, respectively. The complete framework yields a gain of 22.56% in Recall@5 wrt SOTA on our proposed dataset. The code and dataset are available at https://github.com/goyalkaraniit/SymTax, Comment: Accepted in ACL 2024

Large-scale distributed training in production datacenters constitutes a challenging workload bottlenecked by network communication. In response, both major industry players (e.g., Ultra Ethernet Consortium) and parts of academia have surprisingly, and almost unanimously, agreed that packet spraying is necessary to improve the performance of large-scale distributed training workloads. In this paper, we challenge this prevailing belief and pose the question: How close can a singlepath transport approach an optimal multipath transport? We demonstrate that singlepath transport (from a NIC's perspective) is sufficient and can perform nearly as well as an ideal multipath transport with packet spraying, particularly in the context of distributed training in leaf-spine topologies. Our assertion is based on four key observations about workloads driven by collective communication patterns: (i) flows within a collective start almost simultaneously, (ii) flow sizes are nearly equal, (iii) the completion time of a collective is more crucial than individual flow completion times, and (iv) flows can be split upon arrival. We analytically prove that singlepath transport, using minimal flow splitting (at the application layer), is equivalent to an ideal multipath transport with packet spraying in terms of maximum congestion. Our preliminary evaluations support our claims. This paper suggests an alternative agenda for developing next-generation transport protocols tailored for large-scale distributed training.

The deviations of the recent measurements of the muon magnetic moment and the $W-$boson mass from their SM predictions hint to new physics beyond the SM. In this article, we address the observed discrepancies in the $W$-boson mass and muon anomalous magnetic moment in the Inert Two Higgs Doublet Model (I2HDM) extended by a complex scalar field singlet under the SM gauge group. The model is constrained from the existing LEP data and the measurements of partial decay widths to gauge bosons at LHC. It is shown that a large subset of this constrained parameter space of the model can simultaneously accommodate the $W$-boson mass and also explain the muon $g-2$ anomaly., Comment: 15 pages, 5 figures

Recent works have shown that visual pretraining on egocentric datasets using masked autoencoders (MAE) can improve generalization for downstream robotics tasks. However, these approaches pretrain only on 2D images, while many robotics applications require 3D scene understanding. In this work, we propose 3D-MVP, a novel approach for 3D multi-view pretraining using masked autoencoders. We leverage Robotic View Transformer (RVT), which uses a multi-view transformer to understand the 3D scene and predict gripper pose actions. We split RVT's multi-view transformer into visual encoder and action decoder, and pretrain its visual encoder using masked autoencoding on large-scale 3D datasets such as Objaverse. We evaluate 3D-MVP on a suite of virtual robot manipulation tasks and demonstrate improved performance over baselines. We also show promising results on a real robot platform with minimal finetuning. Our results suggest that 3D-aware pretraining is a promising approach to improve sample efficiency and generalization of vision-based robotic manipulation policies. We will release code and pretrained models for 3D-MVP to facilitate future research. Project site: https://jasonqsy.github.io/3DMVP

In this study, we address the challenge of obstacle avoidance for Unmanned Aerial Vehicles (UAVs) through an innovative composite imitation learning approach that combines Proximal Policy Optimization (PPO) with Behavior Cloning (BC) and Generative Adversarial Imitation Learning (GAIL), enriched by the integration of ray-tracing techniques. Our research underscores the significant role of ray-tracing in enhancing obstacle detection and avoidance capabilities. Moreover, we demonstrate the effectiveness of incorporating GAIL in coordinating the flight paths of two UAVs, showcasing improved collision avoidance capabilities. Extending our methodology, we apply our combined PPO, BC, GAIL, and ray-tracing framework to scenarios involving four UAVs, illustrating its scalability and adaptability to more complex scenarios. The findings indicate that our approach not only improves the reliability of basic PPO based obstacle avoidance but also paves the way for advanced autonomous UAV operations in crowded or dynamic environments.

Creating human-like large language model (LLM) agents is crucial for faithful social simulation. Having LLMs role-play based on demographic information sometimes improves human likeness but often does not. This study assessed whether LLM alignment with human behavior can be improved by integrating information from empirically-derived human belief networks. Using data from a human survey, we estimated a belief network encompassing 18 topics loading on two non-overlapping latent factors. We then seeded LLM-based agents with an opinion on one topic, and assessed the alignment of its expressed opinions on remaining test topics with corresponding human data. Role-playing based on demographic information alone did not align LLM and human opinions, but seeding the agent with a single belief greatly improved alignment for topics related in the belief network, and not for topics outside the network. These results suggest a novel path for human-LLM belief alignment in work seeking to simulate and understand patterns of belief distributions in society.

Synthetic long-context LLM benchmarks (e.g., "needle-in-the-haystack") test only surface-level retrieval capabilities, but how well can long-context LLMs retrieve, synthesize, and reason over information across book-length inputs? We address this question by creating NoCha, a dataset of 1,001 minimally different pairs of true and false claims about 67 recently-published English fictional books, written by human readers of those books. In contrast to existing long-context benchmarks, our annotators confirm that the largest share of pairs in NoCha require global reasoning over the entire book to verify. Our experiments show that while human readers easily perform this task, it is enormously challenging for all ten long-context LLMs that we evaluate: no open-weight model performs above random chance (despite their strong performance on synthetic benchmarks), while GPT-4o achieves the highest accuracy at 55.8%. Further analysis reveals that (1) on average, models perform much better on pairs that require only sentence-level retrieval vs. global reasoning; (2) model-generated explanations for their decisions are often inaccurate even for correctly-labeled claims; and (3) models perform substantially worse on speculative fiction books that contain extensive world-building. The methodology proposed in NoCha allows for the evolution of the benchmark dataset and the easy analysis of future models., Comment: preprint, 37 pages

Despite numerous proposals investigating various properties of accelerated detectors in different settings, detecting the Unruh effect remains challenging due to the typically weak signal at achievable accelerations. For an atom with frequency gap $\omega_0$, accelerated in free space, significant acceleration-induced modification of properties like transition rates and radiative energy shifts requires accelerations of the order of $\omega_0 c$. In this paper, we make the case for a suitably modified density of field states to be complemented by a judicious selection of the system property to be monitored. We study the radiative energy-level shift in inertial and uniformly accelerated atoms coupled to a massless quantum scalar field inside a cylindrical cavity. Uniformly accelerated atoms experience thermal correlations in the inertial vacuum, and the radiative shifts are expected to respond accordingly. We show that the noninertial contribution to the energy shift can be isolated and significantly enhanced relative to the inertial contribution by suitably modifying the density of field modes inside a cylindrical cavity. Moreover, we demonstrate that monitoring the radiative energy shift, as compared to transition rates, allows us to reap a stronger purely-noninertial signal. We find that a purely-noninertial radiative shift as large as 50 times the inertial energy shift can be obtained at small, experimentally achievable accelerations ($ a \sim 10^{-9} \omega_{0} c$) if the cavity's radius $R$ is specified with a relative precision of $\delta R/R_{0} \sim 10^{-7}$. Given that radiative shifts for inertial atoms have already been measured with high accuracy, we argue that the radiative energy-level shift is a promising observable for detecting Unruh thermality with current technology., Comment: 8+7 pages, 4 figures

It is a common practice in natural language processing to pre-train a single model on a general domain and then fine-tune it for downstream tasks. However, when it comes to Large Language Models, fine-tuning the entire model can be computationally expensive, resulting in very intensive energy consumption. As a result, several Parameter Efficient Fine-Tuning (PEFT) approaches were recently proposed. One of the most popular approaches is low-rank adaptation (LoRA), where the key insight is decomposing the update weights of the pre-trained model into two low-rank matrices. However, the proposed approaches either use the same rank value across all different weight matrices, which has been shown to be a sub-optimal choice, or do not use any quantization technique, one of the most important factors when it comes to a model's energy consumption. In this work, we propose Bayesian-LoRA which approaches low-rank adaptation and quantization from a Bayesian perspective by employing a prior distribution on both quantization levels and rank values. As a result, B-LoRA is able to fine-tune a pre-trained model on a specific downstream task, finding the optimal rank values and quantization levels for every low-rank matrix. We validate the proposed model by fine-tuning a pre-trained DeBERTaV3 on the GLUE benchmark. Moreover, we compare it to relevant baselines and present both qualitative and quantitative results, showing how the proposed approach is able to learn optimal-rank quantized matrices. B-LoRA performs on par with or better than the baselines while reducing the total number of bit operations by roughly 70% compared to the baseline methods.

Unclonable cryptography utilizes the principles of quantum mechanics to addresses cryptographic tasks that are impossible classically. We introduce a novel unclonable primitive in the context of secret sharing, called unclonable secret sharing (USS). In a USS scheme, there are $n$ shareholders, each holding a share of a classical secret represented as a quantum state. They can recover the secret once all parties (or at least $t$ parties) come together with their shares. Importantly, it should be infeasible to copy their own shares and send the copies to two non-communicating parties, enabling both of them to recover the secret. Our work initiates a formal investigation into the realm of unclonable secret sharing, shedding light on its implications, constructions, and inherent limitations. ** Connections: We explore the connections between USS and other quantum cryptographic primitives such as unclonable encryption and position verification, showing the difficulties to achieve USS in different scenarios. **Limited Entanglement: In the case where the adversarial shareholders do not share any entanglement or limited entanglement, we demonstrate information-theoretic constructions for USS. **Large Entanglement: If we allow the adversarial shareholders to have unbounded entanglement resources (and unbounded computation), we prove that unclonable secret sharing is impossible. On the other hand, in the quantum random oracle model where the adversary can only make a bounded polynomial number of queries, we show a construction secure even with unbounded entanglement. Furthermore, even when these adversaries possess only a polynomial amount of entanglement resources, we establish that any unclonable secret sharing scheme with a reconstruction function implementable using Cliffords and logarithmically many T-gates is also unattainable.

In this work, we study how to build a robotic system that can solve multiple 3D manipulation tasks given language instructions. To be useful in industrial and household domains, such a system should be capable of learning new tasks with few demonstrations and solving them precisely. Prior works, like PerAct and RVT, have studied this problem, however, they often struggle with tasks requiring high precision. We study how to make them more effective, precise, and fast. Using a combination of architectural and system-level improvements, we propose RVT-2, a multitask 3D manipulation model that is 6X faster in training and 2X faster in inference than its predecessor RVT. RVT-2 achieves a new state-of-the-art on RLBench, improving the success rate from 65% to 82%. RVT-2 is also effective in the real world, where it can learn tasks requiring high precision, like picking up and inserting plugs, with just 10 demonstrations. Visual results, code, and trained model are provided at: https://robotic-view-transformer-2.github.io/., Comment: Accepted to RSS 2024

In an ever-expanding world of domain-specific knowledge, the increasing complexity of consuming, and storing information necessitates the generation of summaries from large information repositories. However, every persona of a domain has different requirements of information and hence their summarization. For example, in the healthcare domain, a persona-based (such as Doctor, Nurse, Patient etc.) approach is imperative to deliver targeted medical information efficiently. Persona-based summarization of domain-specific information by humans is a high cognitive load task and is generally not preferred. The summaries generated by two different humans have high variability and do not scale in cost and subject matter expertise as domains and personas grow. Further, AI-generated summaries using generic Large Language Models (LLMs) may not necessarily offer satisfactory accuracy for different domains unless they have been specifically trained on domain-specific data and can also be very expensive to use in day-to-day operations. Our contribution in this paper is two-fold: 1) We present an approach to efficiently fine-tune a domain-specific small foundation LLM using a healthcare corpus and also show that we can effectively evaluate the summarization quality using AI-based critiquing. 2) We further show that AI-based critiquing has good concordance with Human-based critiquing of the summaries. Hence, such AI-based pipelines to generate domain-specific persona-based summaries can be easily scaled to other domains such as legal, enterprise documents, education etc. in a very efficient and cost-effective manner.

Employing language models to generate explanations for an incoming implicit hate post is an active area of research. The explanation is intended to make explicit the underlying stereotype and aid content moderators. The training often combines top-k relevant knowledge graph (KG) tuples to provide world knowledge and improve performance on standard metrics. Interestingly, our study presents conflicting evidence for the role of the quality of KG tuples in generating implicit explanations. Consequently, simpler models incorporating external toxicity signals outperform KG-infused models. Compared to the KG-based setup, we observe a comparable performance for SBIC (LatentHatred) datasets with a performance variation of +0.44 (+0.49), +1.83 (-1.56), and -4.59 (+0.77) in BLEU, ROUGE-L, and BERTScore. Further human evaluation and error analysis reveal that our proposed setup produces more precise explanations than zero-shot GPT-3.5, highlighting the intricate nature of the task., Comment: 17 Pages, 5 Figures, 13 Tables, ACL Findings 2024

The growing integration of distributed energy resources (DERs) into the power grid necessitates an effective coordination strategy to maximize their benefits. Acting as an aggregator of DERs, a virtual power plant (VPP) facilitates this coordination, thereby amplifying their impact on the transmission level of the power grid. Further, a demand response program enhances the scheduling approach by managing the energy demands in parallel with the uncertain energy outputs of the DERs. This work presents a stochastic incentive-based demand response model for the scheduling operation of VPP comprising solar-powered generating stations, battery swapping stations, electric vehicle charging stations, and consumers with controllable loads. The work also proposes a priority mechanism to consider the individual preferences of electric vehicle users and consumers with controllable loads. The scheduling approach for the VPP is framed as a multi-objective optimization problem, normalized using the utopia-tracking method. Subsequently, the normalized optimization problem is transformed into a stochastic formulation to address uncertainties in energy demand from charging stations and controllable loads. The proposed VPP scheduling approach is addressed on a 33-node distribution system simulated using MATLAB software, which is further validated using a real-time digital simulator., Comment: 11 pages, 8 figures, submitted to IEEE Transactions on Industry Applications for potential publication

The inner workings of neural networks can be better understood if we can fully decipher the information encoded in neural activations. In this paper, we argue that this information is embodied by the subset of inputs that give rise to similar activations. Computing such subsets is nontrivial as the input space is exponentially large. We propose InversionView, which allows us to practically inspect this subset by sampling from a trained decoder model conditioned on activations. This helps uncover the information content of activation vectors, and facilitates understanding of the algorithms implemented by transformer models. We present four case studies where we investigate models ranging from small transformers to GPT-2. In these studies, we demonstrate the characteristics of our method, show the distinctive advantages it offers, and provide causally verified circuits., Comment: ICML 2024 Mechanistic Interpretability Workshop oral

We are beginning to see progress in language model assisted scientific discovery. Motivated by the use of LLMs as a general scientific assistant, this paper assesses the domain knowledge of LLMs through its understanding of different mathematical skills required to solve problems. In particular, we look at not just what the pre-trained model already knows, but how it learned to learn from information during in-context learning or instruction-tuning through exploiting the complex knowledge structure within mathematics. Motivated by the Neural Tangent Kernel (NTK), we propose \textit{NTKEval} to assess changes in LLM's probability distribution via training on different kinds of math data. Our systematic analysis finds evidence of domain understanding during in-context learning. By contrast, certain instruction-tuning leads to similar performance changes irrespective of training on different data, suggesting a lack of domain understanding across different skills.

Metric distortion in social choice provides a framework for assessing how well voting rules minimize social cost in scenarios where voters and candidates exist in a shared metric space, with voters submitting rankings and the rule outputting a single winner. We expand this framework to include probabilistic voting. Our extension encompasses a broad range of probability functions, including widely studied models like Plackett-Luce (PL) and Bradley-Terry, and a novel "pairwise quantal voting" model inspired by quantal response theory. We demonstrate that distortion results under probabilistic voting better correspond with conventional intuitions regarding popular voting rules such as Plurality, Copeland, and Random Dictator (RD) than those under deterministic voting. For example, in the PL model with candidate strength inversely proportional to the square of their metric distance, we show that Copeland's distortion is at most 2, whereas that of RD is $\Omega(\sqrt{m})$ in large elections, where $m$ is the number of candidates. This contrasts sharply with the classical model, where RD beats Copeland with a distortion of 3 versus 5 [1].

The ubiquity of "peas-in-a-pod" architectural patterns and the existence of the radius valley each present a striking population-level trend for planets with $R_{p} \leq 4 R_{\oplus}$ that serves to place powerful constraints on the formation and evolution of these subgiant worlds. As it has yet to be determined whether the strength of this peas-in-a-pod uniformity differs on either side of the radius valley, we separately assess the architectures of systems containing only small ($R_{p} \leq 1.6 R_{\oplus}$), rocky planets from those harboring only intermediate-size ($1.6 R_{\oplus} < R_{p} \leq 4 R_{\oplus}$), volatile-rich worlds to perform a novel statistical comparison of intra-system planetary uniformity across compositionally distinct regimes. We find that, compared to their volatile-rich counterparts, rocky systems are less uniform in mass ($2.6\sigma$), but more uniform in size ($4.0\sigma$) and spacing ($3.0\sigma$). We provide further statistical validation for these results, demonstrating that they are not substantially influenced by the presence of mean motion resonances, low-mass host stars, alternative bulk compositional assumptions, sample size effects, or detection biases. We also obtain tentative evidence ($>2 \sigma$ significance) that the enhanced size uniformity of rocky systems is dominated by the presence of super-Earths ($1 R_{\oplus} \leq R_{p} \leq 1.6 R_{\oplus}$), while their enhanced mass diversity is driven by the presence of sub-Earth ($R_{p} < 1 R_{\oplus}$) worlds., Comment: Accepted to ApJ Letters (May 2024). 17 pages (including 3 for Appendix), 4 figures, 3 tables