Searcharxiv⌕ Search

arXiv subjects

Leonel Aguilar

Publications and source records attributed to Leonel Aguilar.

8 recordsLinked to original sources

Kolibri: Color Finds the Place, Refinement Sets the Pose for RGB-D Localization in Colored Point Clouds

An agent or robot in a scanned building must locate its camera in rooms that no model saw during training. Point cloud registration methods can align each RGB-D frame with the building's colored point cloud, accurately but at a high cost. We observe that geometric refinement brings nearby initial poses to the same final pose, so the learned component only needs to find the place. Kolibri does this with a graph network of 55,712 parameters that describes 10 cm cubes of points, called cells, by their color, local shape and neighbors. With the same refinement, it matches GeDi on 96 unseen ScanNet rooms (81.4% against 80.3% within 5 cm and 5°) and on a held-out test of 10 further rooms, in 82 ms per frame on an RTX 2080 Ti. The saved time allows for search without a room label. In one point cloud of 100 rooms, Kolibri loses 1.3 points against knowing the room, while retrieving the room first loses 56.0. In a hospital emergency department scene from a real scan, it localizes 86.5% of the frames rendered from the scan in 384 ms each, where GeDi needs 91 minutes per frame on cubes of the scan for a similar recall. The method needs colors that agree between the frame and the point cloud. On gray frames and across cameras with different color responses, point cloud registration methods keep the lead.

cs.CV↗

Does Your Neural Network Extrapolate? Feature Engineering as Identifiability Bias for OOD Generalization

Deep neural networks can learn salient features from an in-distribution (ID) training window without identifying the data-generating structure needed for out-of-distribution (OOD) generalization. Understanding this limitation requires examining the identifiability of the data-generating process (DGP). From a single training window, the DGP that yields the correct OOD extrapolation is not identifiable: infinitely many DGPs are observationally equivalent within a fixed tolerance on that window yet diverge arbitrarily beyond it, and no ID criterion alone can reliably distinguish them. A structural commitment---the feature map, label map, and model class $(φ, ψ, \mathcal{M})$---specifies the assumed DGP and governs OOD generalization while leaving ID performance essentially unchanged. When architecture, pretraining, data augmentation, input formats, or domain knowledge implicitly supply the correct commitment, the model generalizes beyond the training window. When it cannot infer OOD-relevant structure from ID evidence, it fails to generalize. When the commitment is correct {\em and} identifiable, OOD error vanishes. We demonstrate this mechanism in models trained from scratch and in foundation models across four natural-science settings (chemistry, physics, biology, and histopathology), as well as in a positional-encoding study spanning Transformer, Mamba, and S4D architectures. Finally, a controlled study shows that correct features are necessary but not sufficient for OOD generalization.

cs.LG↗

STRIDE: Automated Evaluation of Text-to-Trajectory Alignment across Diverse Contexts

Language-conditioned trajectory generation is here, but its evaluation has not kept pace. Existing pedestrian trajectory metrics compare trajectories with real-world human data. This does not scale to text-to-trajectory generation across diverse contexts, as collecting human trajectories for every scenario is costly and infeasible. Moreover, pedestrian behavior is heterogeneous and context-dependent, with no single metric as the correct answer, and current evaluation frameworks are not transferable to this domain. These challenges make scalable, reliable evaluation difficult. We introduce STRIDE, the first framework for evaluating context alignment between scenario descriptions and pedestrian trajectories. STRIDE addresses these challenges through three design choices. First, we derive our VRDST evaluation protocol from sociological theories to define a complete evaluation space. Second, it decomposes high-level context into scenario-adaptive behavioral questions. Third, every question is resolved against a deterministic measurement tool library that yields reproducible answers. Together, STRIDE enables verifiable, automated, and scalable evaluation across diverse contexts without requiring human trajectory data. We instantiate STRIDE in the crowd domain as STRIDE-Bench, comprising 1K scenarios, 6K behavioral questions, and 11K measurements with calibrated expected answers across 30 real-world maps. Comprehensive human validations show that STRIDE-Bench is consistent with human behavior and judgment, achieving 80% human agreement. We further evaluate several text-to-trajectory models, finding limited context-alignment capability and persistent challenges in fine-grained context conditioning. We believe that the STRIDE framework provides a first step toward principled evaluation of context-aligned pedestrian trajectory generation.

cs.AI↗

What Should We Freeze? Guarded Freezing: Connectivity Shapes the Fine-Tuning of Pretrained Models

When adapting pre-trained models through fine-tuning, freezing weights alone might not preserve performance, as updates elsewhere can change the inputs to the frozen core, ultimately affecting overall performance. We first analyse the case where a selected frozen core can be isolated and propose removal-value, a capacity-based score that approximates HOPE's removal cost averaged over removal orders. We show that in VGG-8, cutting paths from trainable neurons into a frozen core makes selection using this score useful: $70\%$ frozen preserves $5.22\pm0.51$ percentage points more old-task accuracy than DEFT at similar new-task accuracy. In transformers, shared residual streams leave paths into frozen neurons open. For this case, we derive drift-value, a forward-only proxy for the output disturbance from updating each weight entry under a local update model. In language models, at 40 epochs, this policy exceeds adapted Wanda and RIA freezing scores in settings with substantial retention loss, while its differences from Fisher remain unresolved. After 160 epochs on Qwen2.5-1.5B, it retains $0.0433\pm0.0102$ more than static Fisher. In DINOv3 vision-transformer adaptation to point clouds, drift-value retains $0.440$ image accuracy versus $0.187$ for a random mask of the same count. These results motivate Guarded Freezing: select by removal-value when incoming paths are cut, and by drift-value when they remain.

cs.LG↗

IUMENTA: A generic framework for animal digital twins within the Open Digital Twin Platform

IUMENTA (Latin for livestock) is an innovative software framework designed to construct and simulate digital twins of animals. By leveraging the powerful capability of the Open Digital Twin Platform (ODTP) alongside advanced software sensors, IUMENTA offers researchers a user-friendly tool to seamlessly develop adaptive digital replicas of animal-based processes. This framework establishes a dynamic ecosystem that integrates insights from diverse experiments, consequently enhancing our understanding of animal behavioural and physiological responses. Through real-time tracking of an animal's energy balance. IUMENTA provides valuable insights into metabolic rates, nutritional needs, emotional states, and overall well-being of animals. In this article, we explore the application of the IUMENTA framework in developing a digital twin focused on the animal's energy balance. IUMENTA includes the EnergyTag system, a state-of-the-art wearable software sensor, which facilitates real-time monitoring of energy expenditure, allowing for continuous updates and personalisation of the energy balance digital twin.

cs.OH↗

The Hitchhiker's Guide to Fused Twins: A Review of Access to Digital Twins in situ in Smart Cities

Smart Cities already surround us, and yet they are still incomprehensibly far from directly impacting everyday life. While current Smart Cities are often inaccessible, the experience of everyday citizens may be enhanced with a combination of the emerging technologies Digital Twins (DTs) and Situated Analytics. DTs represent their Physical Twin (PT) in the real world via models, simulations, (remotely) sensed data, context awareness, and interactions. However, interaction requires appropriate interfaces to address the complexity of the city. Ultimately, leveraging the potential of Smart Cities requires going beyond assembling the DT to be comprehensive and accessible. Situated Analytics allows for the anchoring of city information in its spatial context. We advance the concept of embedding the DT into the PT through Situated Analytics to form Fused Twins (FTs). This fusion allows access to data in the location that it is generated in an embodied context that can make the data more understandable. Prototypes of FTs are rapidly emerging from different domains, but Smart Cities represent the context with the most potential for FTs in the future. This paper reviews DTs, Situated Analytics, and Smart Cities as the foundations of FTs. Regarding DTs, we define five components (Physical, Data, Analytical, Virtual, and Connection environments) that we relate to several cognates (i.e., similar but different terms) from existing literature. Regarding Situated Analytics, we review the effects of user embodiment on cognition and cognitive load. Finally, we classify existing partial examples of FTs from the literature and address their construction from Augmented Reality, Geographic Information Systems, Building/City Information Models, and DTs and provide an overview of future direction

cs.CY↗

Experiments as Code: A Concept for Reproducible, Auditable, Debuggable, Reusable, & Scalable Experiments

A common concern in experimental research is the auditability and reproducibility of experiments. Experiments are usually designed, provisioned, managed, and analyzed by diverse teams of specialists (e.g., researchers, technicians and engineers) and may require many resources (e.g. cloud infrastructure, specialized equipment). Even though researchers strive to document experiments accurately, this process is often lacking, making it hard to reproduce them. Moreover, when it is necessary to create a similar experiment, very often we end up "reinventing the wheel" as it is easier to start from scratch than trying to reuse existing work, thus losing valuable embedded best practices and previous experiences. In behavioral studies this has contributed to the reproducibility crisis. To tackle this challenge, we propose the "Experiments as Code" paradigm, where the whole experiment is not only documented but additionally the automation code to provision, deploy, manage, and analyze it is provided. To this end we define the Experiments as Code concept, provide a taxonomy for the components of a practical implementation, and provide a proof of concept with a simple desktop VR experiment that showcases the benefits of its "as code" representation, i.e., reproducibility, auditability, debuggability, reusability, and scalability.

cs.CY↗

How intelligence can change the course of evolution

The effect of phenotypic plasticity on evolution, the so-called Baldwin effect, has been studied extensively for more than 100 years. Plasticity is known to influence the speed of evolution towards a specific genetic configuration, but whether it also influences what that genetic configuration is, is still an open question. This question is investigated, in an environment where the distribution of resources follows seasonal cycles, both analytically and experimentally by means of an agent-based model of a foraging task. Individuals can either specialize to foraging only one specific resource type or generalize to foraging all resource types at a low success rate. It is found that the introduction of learning, one instance of phenotypic plasticity, changes what genetic configuration evolves. Specifically, the genome of learning agents evolves a predisposition to adapt quickly to changes in the resource distribution, under the same conditions for which non-learners would evolve a predisposition to maximize the foraging efficiency for a specific resource type. This paper expands the literature at the interface between Biology and Machine Learning by identifying the Baldwin effects in cyclically-changing environments and demonstrating that learning can change the outcome of evolution.

q-bio.PE↗