Search

The current biodiversity loss crisis makes animal monitoring a relevant field of study. In light of this, data collected through monitoring can provide essential insights, and information for decision-making aimed at preserving global biodiversity. Despite the importance of such data, there is a notable scarcity of datasets featuring videos of birds, and none of the existing datasets offer detailed annotations of bird behaviors in video format. In response to this gap, our study introduces the first fine-grained video dataset specifically designed for bird behavior detection and species classification. This dataset addresses the need for comprehensive bird video datasets and provides detailed data on bird actions, facilitating the development of deep learning models to recognize these, similar to the advancements made in human action recognition. The proposed dataset comprises 178 videos recorded in Spanish wetlands, capturing 13 different bird species performing 7 distinct behavior classes. In addition, we also present baseline results using state of the art models on two tasks: bird behavior recognition and species classification.

Large language models must balance their weight-encoded knowledge with in-context information from prompts to generate accurate responses. This paper investigates this interplay by analyzing how models of varying capacities within the same family handle intentionally misleading in-context information. Our experiments demonstrate that larger models exhibit higher resilience to deceptive prompts, showcasing an advanced ability to interpret and integrate prompt information with their internal knowledge. Furthermore, we find that larger models outperform smaller ones in following legitimate instructions, indicating that their resilience is not due to disregarding in-context information. We also show that this phenomenon is likely not a result of memorization but stems from the models' ability to better leverage implicit task-relevant information from the prompt alongside their internally stored knowledge.

Multimodal AI has the potential to significantly enhance document-understanding tasks, such as processing receipts, understanding workflows, extracting data from documents, and summarizing reports. Code generation tasks that require long-structured outputs can also be enhanced by multimodality. Despite this, their use in commercial applications is often limited due to limited access to training data and restrictive licensing, which hinders open access. To address these limitations, we introduce BigDocs-7.5M, a high-quality, open-access dataset comprising 7.5 million multimodal documents across 30 tasks. We use an efficient data curation process to ensure our data is high-quality and license-permissive. Our process emphasizes accountability, responsibility, and transparency through filtering rules, traceable metadata, and careful content analysis. Additionally, we introduce BigDocs-Bench, a benchmark suite with 10 novel tasks where we create datasets that reflect real-world use cases involving reasoning over Graphical User Interfaces (GUI) and code generation from images. Our experiments show that training with BigDocs-Bench improves average performance up to 25.8% over closed-source GPT-4o in document reasoning and structured output tasks such as Screenshot2HTML or Image2Latex generation. Finally, human evaluations showed a preference for outputs from models trained on BigDocs over GPT-4o. This suggests that BigDocs can help both academics and the open-source community utilize and improve AI tools to enhance multimodal capabilities and document reasoning. The project is hosted at https://bigdocs.github.io ., Comment: The project is hosted at https://bigdocs.github.io

NISQ (Noisy Intermediate-Scale Quantum) era constraints, high sensitivity to noise and limited qubit count, impose significant barriers on the usability of QPUs (Quantum Process Units) capabilities. To overcome these challenges, researchers are exploring methods to maximize the utility of existing QPUs despite their limitations. Building upon the idea that the execution of a quantum circuit's shots needs not to be treated as a singular monolithic unit, we propose a methodological framework, termed shot-wise, which enables the distribution of shots for a single circuit across multiple QPUs. Our framework features customizable policies to adapt to various scenarios. Additionally, it introduces a calibration method to pre-evaluate the accuracy and reliability of each QPU's output before the actual distribution process and an incremental execution mechanism for dynamically managing the shot allocation and policy updates. Such an approach enables flexible and fine-grained management of the distribution process, taking into account various user-defined constraints and (contrasting) objectives. Experimental findings show that while these strategies generally do not exceed the best individual QPU results, they maintain robustness and align closely with average outcomes. Overall, the shot-wise methodology improves result stability and often outperforms single QPU runs, offering a flexible approach to managing variability in quantum computing., Comment: 22 pages, 7 figures

In today's digitally driven world, dialogue systems play a pivotal role in enhancing user interactions, from customer service to virtual assistants. In these dialogues, it is important to identify user's goals automatically to resolve their needs promptly. This has necessitated the integration of models that perform Intent Detection. However, users' intents are diverse and dynamic, making it challenging to maintain a fixed set of predefined intents. As a result, a more practical approach is to develop a model capable of identifying new intents as they emerge. We address the challenge of Intent Discovery, an area that has drawn significant attention in recent research efforts. Existing methods need to train on a substantial amount of data for correctly identifying new intents, demanding significant human effort. To overcome this, we introduce IntentGPT, a novel training-free method that effectively prompts Large Language Models (LLMs) such as GPT-4 to discover new intents with minimal labeled data. IntentGPT comprises an \textit{In-Context Prompt Generator}, which generates informative prompts for In-Context Learning, an \textit{Intent Predictor} for classifying and discovering user intents from utterances, and a \textit{Semantic Few-Shot Sampler} that selects relevant few-shot examples and a set of known intents to be injected into the prompt. Our experiments show that IntentGPT outperforms previous methods that require extensive domain-specific data and fine-tuning, in popular benchmarks, including CLINC and BANKING, among others., Comment: ICLR 2024 Workshop on LLM Agents

The action of the subgroup $\operatorname{G}_2$ of $\operatorname{SO}(7)$ (resp.\ $\operatorname{Spin}(7)$ of $\operatorname{SO}(8)$) on the Grassmannian space $M=\frac{\operatorname{SO}(7)}{\operatorname{SO}(5)\times\operatorname{SO}(2)}$ (resp.\ $M=\frac{\operatorname{SO}(8)}{\operatorname{SO}(5)\times\operatorname{SO}(3)}$) is still transitive. We prove that the spectrum (i.e.\ the collection of eigenvalues of its Laplace-Beltrami operator) of a symmetric metric $g_0$ on $M$ coincides with the spectrum of a $\operatorname{G}_2$-invariant (resp.\ $\operatorname{Spin}(7)$-invariant) metric $g$ on $M$ only if $g_0$ and $g$ are isometric. As a consequence, each non-flat compact irreducible symmetric space of non-group type is spectrally unique among the family of all currently known homogeneous metrics on its underlying differentiable manifold.

Random Forest (RF) and Gradient Boosting Regression Trees (GBRT) regression models along with three cheminformatics data sets (RDkit, Mordred, Morgan) have been used to predict the power conversion efficiency (PCE) of organic solar cells (OSCs). The data consists of cheinformatics descriptors of the electron donor used in 433 OSCs for which the experimental PCE (target variable) is reported in the literature. The donor is either a polymer or a small organic molecule, and the acceptor the fullerene derivatives PCBM or PC71BM. Unlike previous methods, our ML approach considers the type of donor and the acceptor by adding four extra donor's features using the one-hot encoder tool. It is demonstrated that this additional information improves the prediction performance up to 34%. We have also exploited this feature to theoretically forecast the PCE of new OSCs by evaluating the ML model for a different acceptor. It is predicted that more than 50% of the OCSs obtained by exchanging the acceptor would have higher experimental PCE. The prediction accuracy of a given ML approach is analyzed for different PCE intervals. RF using RDkit descriptors resulted in the best ML approach with a Pearson's correlation coefficient for the training and testing sets equal to 0.96 and 0.62, respectively., Comment: 12 pages, 4 Figures, 1 Table

Property inheritance -- a phenomenon where novel properties are projected from higher level categories (e.g., birds) to lower level ones (e.g., sparrows) -- provides a unique window into how humans organize and deploy conceptual knowledge. It is debated whether this ability arises due to explicitly stored taxonomic knowledge vs. simple computations of similarity between mental representations. How are these mechanistic hypotheses manifested in contemporary language models? In this work, we investigate how LMs perform property inheritance with behavioral and causal representational analysis experiments. We find that taxonomy and categorical similarities are not mutually exclusive in LMs' property inheritance behavior. That is, LMs are more likely to project novel properties from one category to the other when they are taxonomically related and at the same time, highly similar. Our findings provide insight into the conceptual structure of language models and may suggest new psycholinguistic experiments for human subjects.

This paper examines the impact of electric vehicle (EV) charging stations on the capacity of distribution feeders and transformers within the electric grid at California State University Northridge (CSUN). With the increasing adoption of both residential and commercial EVs and the rapid expansion of EV charging infrastructure, it is critical to evaluate the potential overloading effects of intensive EV charging on power distribution systems. This research assesses the impact of EV charging on the operation of CSUN's electric grid, identifying potential overload risks under projected EV adoption scenarios. Detailed simulations and analyses are conducted to quantify the extent of these impacts, focusing on various levels of EV penetration and charging patterns. The study also explores the impact of distributed generation on reducing the stress incurred by EV loads. The findings provide essential insights for utility companies, highlighting the need for strategic upgrades to distribution systems. These insights will help in developing robust strategies for both current operations and future planning to accommodate growing EV charging demands, ensuring grid stability and reliability in the face of increasing electrification of the transportation sector. The modeled CSUN electric grid can be used as a benchmark to study the impact of EV loads in dense areas on various parameters of the grid., Comment: 7 pages

Chain-of-thought (CoT) via prompting is the de facto method for eliciting reasoning capabilities from large language models (LLMs). But for what kinds of tasks is this extra ``thinking'' really helpful? To analyze this, we conducted a quantitative meta-analysis covering over 100 papers using CoT and ran our own evaluations of 20 datasets across 14 models. Our results show that CoT gives strong performance benefits primarily on tasks involving math or logic, with much smaller gains on other types of tasks. On MMLU, directly generating the answer without CoT leads to almost identical accuracy as CoT unless the question or model's response contains an equals sign, indicating symbolic operations and reasoning. Following this finding, we analyze the behavior of CoT on these problems by separating planning and execution and comparing against tool-augmented LLMs. Much of CoT's gain comes from improving symbolic execution, but it underperforms relative to using a symbolic solver. Our results indicate that CoT can be applied selectively, maintaining performance while saving inference costs. Furthermore, they suggest a need to move beyond prompt-based CoT to new paradigms that better leverage intermediate computation across the whole range of LLM applications., Comment: Swapped column names for Table 7 and 8 in the appendix. Fixed the prompt for SocialIQA; results in figures and tables are updated (no major differences, but the prompt is now correct)

Data analytics is essential for extracting valuable insights from data that can assist organizations in making effective decisions. We introduce InsightBench, a benchmark dataset with three key features. First, it consists of 100 datasets representing diverse business use cases such as finance and incident management, each accompanied by a carefully curated set of insights planted in the datasets. Second, unlike existing benchmarks focusing on answering single queries, InsightBench evaluates agents based on their ability to perform end-to-end data analytics, including formulating questions, interpreting answers, and generating a summary of insights and actionable steps. Third, we conducted comprehensive quality assurance to ensure that each dataset in the benchmark had clear goals and included relevant and meaningful questions and analysis. Furthermore, we implement a two-way evaluation mechanism using LLaMA-3 as an effective, open-source evaluator to assess agents' ability to extract insights. We also propose AgentPoirot, our baseline data analysis agent capable of performing end-to-end data analytics. Our evaluation on InsightBench shows that AgentPoirot outperforms existing approaches (such as Pandas Agent) that focus on resolving single queries. We also compare the performance of open- and closed-source LLMs and various evaluation strategies. Overall, this benchmark serves as a testbed to motivate further development in comprehensive automated data analytics.

We introduce a machine learning (ML) supervised model function that is inspired by the variational principle of physics. This ML hypothesis evolutionary method, termed ML-Omega, allows us to go from data to differential equation(s) underlying the physical (chemical, engineering, etc.) phenomena the data are derived from. The fundamental equations of physics can be derived from this ML-Omega evolutionary method when provided the proper training data. By training the ML-Omega model function with only three hydrogen-like atom energies, the method can find Schr\"odinger's exact functional and, from it, Schr\"odinger's fundamental equation. Then, in the field of density functional theory (DFT), when the model function is trained with the energies from the known Thomas-Fermi (TF) formula E = -0.7687Z^7/3, it correctly finds the exact TF functional. Finally, the method is applied to find a local orbital-free (OF) functional expression of the independent electron kinetic energy functional Ts based on the gamma-TF-lambda-vW model. By considering the theoretical energies of only 5 atoms (He, Be, Ne, Mg, Ar) as the training set, the evolutionary ML-Omega method finds an ML-Omega-OF-DFT local Ts functional (gamma-TF-lambda-vW (0.964, 1/4)) that outperforms all the OF- DFT functionals of a representative group. Moreover, our ML-Omega-OF functional overcomes the LDA's and some local GGA-DFT's functionals' difficulty to describe the stretched bond region at the correct spin configuration of diatomic molecules. Although our evolutionary ML-Omega model function can work without an explicit prior-form functional, by using the techniques of symbolic regression, in this work we exploit prior-form functional expressions to make the training process faster in the example problems presented here.

CyberShake (CS) is a high-performance computing workflow for Probabilistic Seismic Hazard Assessment (PSHA) developed by the Statewide California Earthquake Center. Here, we employ CS to generate a set of 2103 fault ruptures and simulate the corresponding two horizontal velocity components time histories of ground motion (GM) on a 5-km grid of 625 stations in Southwest Iceland (SI). The ruptures were defined on a new synthetic time-independent 500-year catalog consisting of 223 hypothetical finite-fault sources of 5-7, generated using a new physics-based bookshelf fault system model in the SI transform zone. This fault system model and rupture realizations enable the CS time-independent physics-based approach to PSHA in the region. The study aims to migrate CS to SI and validate its kinematic fault rupture, anelastic wave propagation and ground motion simulations. Toward this goal, we use CS to generate multiple finite-fault rupture variations for each hypothetical fault. CS exploits seismic reciprocity for wave propagation by computing Strain Green Tensors along fault planes, which in turn are convolved with rupture models to generate GM seismograms. For each GM recording station, every adjoint simulation uses a 0-1 Hz Gaussian point source polarized along one horizontal grid direction. Comparison of the results in the form of rotation-invariant synthetic pseudo-acceleration spectral response values at 2, 3 and 5 sec periods are in very good agreement with the Icelandic strong-motion dataset, and a suite of new empirical Bayesian ground motion prediction equations (GMPEs). The vast majority of the CS results fall within one standard deviation of the mean GMPE predictions, previously estimated for the area. Importantly, at large magnitudes for which no data exists in Iceland, the CS dataset may play an important role in constraining the GMPEs for future applications., Comment: 23 pages and 7 figures

The great expansion in offshore power plants is raising the concern regarding the cumulative effect of the electromagnetic field emissions caused by submarine power cables. In this sense, owners are required to predict these emissions during the permitting and consenting process of new power plants. This is a challenging task, especially in the case of HVAC three-core armored cables due to their complex geometry. Customarily, 2D approaches based on the finite element method (FEM) have been employed for evaluating the magnetic field emissions caused by these cables. However, inaccurate results are obtained since the phase conductors and armor twisting is omitted. This work develops, for the first time in the literature, an in-depth analysis of the magnetic field caused by this type of cable through an ultra-shortened 3D-FEM model, which is also faced to experimental measurements taken on an actual 132 kV, 800 mm2 three-core armored cable. Relevant conclusions are derived regarding the impact of the cable design on the magnetic field emissions, including material properties, as well as single and double-layer armors, presenting the proposed model not only as a valuable tool for predicting purposes, but also for optimizing cable design in terms of magnetic field emissions.

Scalable Vector Graphics (SVGs) are vital for modern image rendering due to their scalability and versatility. Previous SVG generation methods have focused on curve-based vectorization, lacking semantic understanding, often producing artifacts, and struggling with SVG primitives beyond path curves. To address these issues, we introduce StarVector, a multimodal large language model for SVG generation. It performs image vectorization by understanding image semantics and using SVG primitives for compact, precise outputs. Unlike traditional methods, StarVector works directly in the SVG code space, leveraging visual understanding to apply accurate SVG primitives. To train StarVector, we create SVG-Stack, a diverse dataset of 2M samples that enables generalization across vectorization tasks and precise use of primitives like ellipses, polygons, and text. We address challenges in SVG evaluation, showing that pixel-based metrics like MSE fail to capture the unique qualities of vector graphics. We introduce SVG-Bench, a benchmark across 10 datasets, and 3 tasks: Image-to-SVG, Text-to-SVG generation, and diagram generation. Using this setup, StarVector achieves state-of-the-art performance, producing more compact and semantically rich SVGs.

We prove that the irreducible symmetric space of complex structures on $\mathbb R^{2n}$ (resp.\ quaternionic structures on $\mathbb C^{2n}$) is spectrally unique within a $2$-parameter (resp.\ $3$-parameter) family of homogeneous metrics on the underlying differentiable manifold. Such families are strong candidates to contain all homogeneous metrics admitted on the corresponding manifolds. The main tool in the proof is an explicit expression for the smallest positive eigenvalue of the Laplace-Beltrami operator associated to each homogeneous metric involved. As a second consequence of this expression, we prove that any non-symmetric Einstein metric in the homogeneous families mentioned above are $\nu$-unstable.

Decades of research indicate that emotion recognition is more effective when drawing information from multiple modalities. But what if some modalities are sometimes missing? To address this problem, we propose a novel Transformer-based architecture for recognizing valence and arousal in a time-continuous manner even with missing input modalities. We use a coupling of cross-attention and self-attention mechanisms to emphasize relationships between modalities during time and enhance the learning process on weak salient inputs. Experimental results on the Ulm-TSST dataset show that our model exhibits an improvement of the concordance correlation coefficient evaluation of 37% when predicting arousal values and 30% when predicting valence values, compared to a late-fusion baseline approach.

We present Lil-Bevo, our submission to the BabyLM Challenge. We pretrained our masked language models with three ingredients: an initial pretraining with music data, training on shorter sequences before training on longer ones, and masking specific tokens to target some of the BLiMP subtasks. Overall, our baseline models performed above chance, but far below the performance levels of larger LLMs trained on more data. We found that training on short sequences performed better than training on longer sequences.Pretraining on music may help performance marginally, but, if so, the effect seems small. Our targeted Masked Language Modeling augmentation did not seem to improve model performance in general, but did seem to help on some of the specific BLiMP tasks that we were targeting (e.g., Negative Polarity Items). Training performant LLMs on small amounts of data is a difficult but potentially informative task. While some of our techniques showed some promise, more work is needed to explore whether they can improve performance more than the modest gains here. Our code is available at https://github.com/venkatasg/Lil-Bevo and out models at https://huggingface.co/collections/venkatasg/babylm-653591cdb66f4bf68922873a, Comment: Proceedings of the BabyLM Challenge

Understanding when two pieces of text convey the same information is a goal touching many subproblems in NLP, including textual entailment and fact-checking. This problem becomes more complex when those two pieces of text are in different languages. Here, we introduce X-PARADE (Cross-lingual Paragraph-level Analysis of Divergences and Entailments), the first cross-lingual dataset of paragraph-level information divergences. Annotators label a paragraph in a target language at the span level and evaluate it with respect to a corresponding paragraph in a source language, indicating whether a given piece of information is the same, new, or new but can be inferred. This last notion establishes a link with cross-language NLI. Aligned paragraphs are sourced from Wikipedia pages in different languages, reflecting real information divergences observed in the wild. Armed with our dataset, we investigate a diverse set of approaches for this problem, including token alignment from machine translation, textual entailment methods that localize their decisions, and prompting LLMs. Our results show that these methods vary in their capability to handle inferable information, but they all fall short of human performance., Comment: To be published in NAACL 2024

This paper provides an overview of the approach we used as team ISISTANITOS for the ACM RecSys Challenge 2023. The competition was organized by ShareChat, and involved predicting the probability of a user clicking an app ad and/or installing an app, to improve deep funnel optimization and a special focus on user privacy. Our proposed method inferring the probabilities of clicking and installing as two different, but related tasks. Hence, the model engineers a specific set of features for each task and a set of shared features. Our model is called Weighted Multi-Level Feature Factorization because it considers the interaction of different order features, where the order is associated to the depth in a neural network. The prediction for a given task is generated by combining the task specific and shared features on the different levels. Our submission achieved the 11 rank and overall score of 55 in the competition academia-track final results. We release our source code at: https://github.com/knife982000/RecSys2023Challenge

In this paper we present NR-SLAM, a novel non-rigid monocular SLAM system founded on the combination of a Dynamic Deformation Graph with a Visco-Elastic deformation model. The former enables our system to represent the dynamics of the deforming environment as the camera explores, while the later allows us to model general deformations in a simple way. The presented system is able to automatically initialize and extend a map modeled by a sparse point cloud in deforming environments, that is refined with a sliding-window Deformable Bundle Adjustment. This map serves as base for the estimation of the camera motion and deformation and enables us to represent arbitrary surface topologies, overcoming the limitations of previous methods. To assess the performance of our system in challenging deforming scenarios, we evaluate it in several representative medical datasets. In our experiments, NR-SLAM outperforms previous deformable SLAM systems, achieving millimeter reconstruction accuracy and bringing automated medical intervention closer. For the benefit of the community, we make the source code public., Comment: 12 pages, 7 figures, submited to the IEEE Transactions on Robotics (T-RO)

Safety-critical systems such as those in automotive, avionics and space, require appropriate safety measures to avoid silent data corruption upon random hardware errors such as those caused by radiation and other types of electromagnetic interference. Those safety measures must be able to prevent faults from causing the so-called common cause failures (CCFs), which occur when a fault produces identical errors in redundant elements so that comparison fails to detect the errors and a failure arises. The usual solution to avoid CCFs in CPU cores is using lockstep cores, so that two cores execute the same flow of instructions, but with some time staggering so that their state is never identical and faults can only lead to different errors, which are then detectable by means of comparison. This paper extends Gaisler's RISC-V NOEL-V core with lockstep; and presents future prospects for its use and distribution., Comment: Abstract presented at the RISC-V Summit, June 2023, Barcelona (Spain)

The generative modeling landscape has experienced tremendous growth in recent years, particularly in generating natural images and art. Recent techniques have shown impressive potential in creating complex visual compositions while delivering impressive realism and quality. However, state-of-the-art methods have been focusing on the narrow domain of natural images, while other distributions remain unexplored. In this paper, we introduce the problem of text-to-figure generation, that is creating scientific figures of papers from text descriptions. We present FigGen, a diffusion-based approach for text-to-figure as well as the main challenges of the proposed task. Code and models are available at https://github.com/joanrod/figure-diffusion, Comment: Published at ICLR 2023 as a Tiny Paper

Textual entailment models are increasingly applied in settings like fact-checking, presupposition verification in question answering, or summary evaluation. However, these represent a significant domain shift from existing entailment datasets, and models underperform as a result. We propose WiCE, a new fine-grained textual entailment dataset built on natural claim and evidence pairs extracted from Wikipedia. In addition to standard claim-level entailment, WiCE provides entailment judgments over sub-sentence units of the claim, and a minimal subset of evidence sentences that support each subclaim. To support this, we propose an automatic claim decomposition strategy using GPT-3.5 which we show is also effective at improving entailment models' performance on multiple datasets at test time. Finally, we show that real claims in our dataset involve challenging verification and retrieval problems that existing models fail to address., Comment: EMNLP 2023