We built a hybrid multi-agent trading system that combines large language model (LLM) analysts with momentum-based portfolio construction and rule-based execution. Company-news and macroeconomic agents return structured reports to a coordinator that sets stock weights and cash reserves. We evaluate the portfolio engine using its broker-recorded paper account, which operated with deterministic fallbacks, and compare LLM-enabled configurations through retrospective replays. From March 11 to August 28, 2026, the account returned 34.5%, compared with 14.4% for the S&P 500 fund SPY over the same dates. This raw return advantage came with a larger maximum drawdown. Reconstructed policies also finish above matched SPY and QQQ references when assessments are used in the same session or deferred to the next. Design comparisons examine cash authority, selective invocation, and response contracts. A constant-cash control gives a return estimate close to stance-based sizing, and shorter report contracts improve observed parsing but can change recommendations.
This study investigates whether generalized means can achieve forecasting performance comparable to that of TimeGPT for daily emergency call volume prediction. Specifically, two families of generalized means, the H\"older means (HM) and the Lehmer means (LM), are evaluated against fine-tuned TimeGPT across forecasting horizons ranging from one to seven days. Forecasting performance is assessed on a real-world emergency call dataset using multiple error metrics, and Diebold-Mariano test. Experimental results show that HM and LM achieve nearly identical predictive performance across all forecasting horizons. For short-term forecasting, both generalized means outperform TimeGPT, with reductions of mean absolute error (MAE) approximately 1.24\% and 2.48\% at horizons of one and three days, respectively. Conversely, TimeGPT achieves a lower error at the seven-day horizon, with an descresing of MAE by 1.98\%. However, Diebold-Mariano tests reveal no statistically significant differences between the methods at the evaluated horizons. These findings suggest that generalized means can provide competitive forecasting performance compared with a energyvore foundation model.
Digital programmatic advertising operates through complex, multi-hop supply chains. The opacity of these intermediary networks leads to significant efficiency losses, often referred to as the ``ad-tech tax.'' While current industry practices utilize heuristic Supply Path Optimization (SPO) to trim inefficient routes, they fail to optimize the equilibrium of the entire network. This paper extends previous work on Directed Acyclic Graph (DAG) intermediation models by applying them to traceable programmatic supply chains mapped via the OpenRTB SupplyChain object. We introduce multi-attribute utility functions and Mixed-Integer Linear Programming (MILP) to account for substitutable routing (OR nodes) and percentage-based fee structures. Furthermore, we outline a computational methodology and experimental framework using public \texttt{schain} and \texttt{sellers.json} data to benchmark optimal Nash social welfare against current uncoordinated market outcomes.
We study the numerical approximation of Schr\"{o}dinger Bridges in the setting where samples from the target distribution are unavailable and only an unnormalized target density is given. In this regime, the widely used iterative proportional fitting (IPF) procedure (and its variants) often exhibits severe numerical instabilities, stemming from ill-conditioned underlying control problems and from the fact that the optimal solution may lie far from the initialization. To address these issues, we propose a trust region variant of IPF that constrains successive updates to remain at a fixed Kullback-Leibler divergence from the current iterate. This restriction significantly improves numerical stability. Moreover, we show that the resulting method admits an interpretation as a mirror descent scheme with an adaptively chosen step size. We demonstrate the effectiveness of our approach on several challenging sampling tasks.
Quantifying mouse behavior from pose is a foundational tool in neuroscience, ethology, drug discovery, and animal welfare. Self-supervised foundation models are a natural fit, but a foundation model for mouse behavior must handle social behavior - many behaviors that matter (aggression, courtship, mounting, allogrooming) are inherently multi-animal. Existing self-supervised pose models fall short: they encode each animal independently and discard interaction signal, hard-code the animal count, and pretrain one dataset at a time. We introduce MICE (Multi-animal Interaction Context Encoder), a hierarchical foundation model for mouse behavior that pretrains jointly on six mouse-pose corpora spanning one to four mice and 7 to 27 keypoints. MICE combines a hierarchical individual encoder capturing per-mouse kinematics at multiple temporal scales with a Perceiver-style social encoder whose learned latent codebook decouples parameter count from animal count, so a single checkpoint serves any group size without retraining. Across four evaluation protocols spanning six datasets, MICE matches or exceeds prior keypoint-based baselines, and layer-wise probing shows the social stage incrementally improves discriminability beyond the individual encoder. Code, weights, and data are available at https://github.com/eric0708/MICE.
We present Feel My Focus, a gaze-triggered mixed reality (MR) demonstration that enables rapid creation and sharing of lightweight 3D digital twins during remote collaboration. Unlike systems that use gaze primarily as a referential pointer, Feel My Focus treats sustained visual attention as a generative action: a fixation on a physical object initiates single-image capture, attention-guided segmentation, and one-shot 3D reconstruction, allowing a remote partner to inspect and discuss the object in situ. The system integrates gaze-tracking glasses, modular segmentation and reconstruction services, and an MR client for runtime streaming and interaction. This design prioritizes responsiveness and experiential continuity over photometric fidelity. We report observations from a public demonstration, highlighting design trade-offs, failure modes, and user control considerations for deploying gaze-triggered generative systems in real-world collaborative settings.
Data-driven remaining useful life (RUL) estimators for turbofan engines are accurate but opaque, and their internal features typically encode remaining life rather than any specific component's state. This paper presents an interpretable genetic fuzzy tree that predicts RUL from cruise-condition sensor means on N-CMAPSS while exposing a per-component diagnostic layer inspectable rule by rule. The tree follows the engine hierarchy: sensors to component health, then to bounded component damage, spool damage, and RUL, so each level answers a different maintenance question. Training is staged so the prognostic objective cannot reshape the diagnostic layers: health leaves are fitted to physical targets and frozen, a deterministic fuzzy damage map with fleet-referenced thresholds is imposed on each component, and only the aggregation is optimized for RUL. It consumes 13 cruise-mean scalars per cycle rather than a trajectory, which removes any sequence model. On the native split it reaches 8.44 cycles RMSE on the multi-fault subset and 11.51 across all modes with 333 free parameters, 411 total rules, 32 to 12,000 times fewer than compiled baselines. An exhaustive sensor search bounds which components are diagnosable under the tested cruise-mean design, and the rule base is structured to admit formal verification, which this study does not discharge.
When a scanpath descriptor discriminates a task, it is rarely clear what the discrimination rests on. We propose a perturbation profile for trial-level descriptors. Five transformations each destroy one stated property while preserving a listed set of invariants: fixation order, duration-position coupling, the duration distribution, orientation and handedness, and fine spatial structure. The classifier is refit, and the drop in participant-held-out AUC is the pipeline's sensitivity to that property. Because refitting lets the learner use redundant information, this describes the refitted pipeline, not what a fitted classifier used. Intervals come from a participant bootstrap that refits; in simulations at the sample sizes analysed, where the target was zero they excluded zero at a pooled rate of at most 2.5%, and they covered the target in 99.6% of 2720 cases. We profiled seven descriptors on twelve binary problems from three public datasets: reading (SB-SAT, 95 participants), program comprehension (EMIP, up to 159) and video and game viewing (GazeBase, 60). Permuting fixation order raised the AUC of a descriptor built on mean saccade amplitude on three problems (by up to 0.23), because permutation turns amplitude into a measure of spatial spread. A "comprehension" contrast that pooled two programs reached AUC 0.70 with a spread-only baseline and fell to 0.47 within one program: the profile located the confound in stimulus layout. On native versus non-native readers, recurrence was sensitive to order and to jitter at the scale of text lines but not of characters, while a two-feature duration and amplitude descriptor reached the same AUC (0.65). A graph descriptor built on the code's tokens was the strongest non-baseline descriptor on the program contrast (0.78; 0.87 with boosted trees). The interpreted results held under longer runs and three seeds; under a second learner 15 of 16 kept t
Ship design requires coordinating coupled subsystems to satisfy engineering constraints and balance competing performance and cost objectives. Generative artificial intelligence has supported component design, but integrated ship generation must also account for interactions among hull geometry, structure, propulsion, and operation. This paper presents GenISys, a framework for generating integrated systems, demonstrated through early-stage bulk-carrier design. Its dataset contains 217,999 proposed designs, including 16,247 that satisfy the formulated engineering constraints. The proposed simultaneous conditional denoising diffusion probabilistic model (SCDDPM) generates subsystems together, allowing each subsystem model to respond to changes elsewhere during generation. Experiments compare this approach with monolithic whole-ship generation and sequential subsystem generation across five independent training runs per architecture. In a study targeting lower cost per tonne of cargo transported, designs were generated for a required voyage range of 5,000 nautical miles. SCDDPM produced designs satisfying the formulated engineering constraints and meeting a common low-cost target in 55.70% of generation requests, compared with 51.10% for the monolithic model and 47.54% for the sequential model. These correspond to 9.0% and 17.2% more qualifying designs, respectively, while maintaining similar estimated transport costs. These results show how generative artificial intelligence can coordinate subsystems to generate feasible, high-performing integrated systems.
Test-time scaling improves the accuracy of a language model's answers by spending more computation at inference time. Established methods do so either sequentially, revising a single solution over several passes, or in parallel, generating several independent solutions and selecting among them. We present ODSA (Orchestrated Discovery and Self-Aggregation), an LLM-directed scaling framework in which an orchestrator LLM is given existing test-time scaling methods as actions --- discovery of a new independent solution, comparison of existing solutions to build a better one, and acceptance of a proposed answer, which triggers its verification --- and directs stateless workers to carry them out. The orchestrator does not solve the problem itself: it allocates computation, feeds the results of earlier actions into later ones, and takes every action from a single context, so that each decision is informed by everything tried before. On Humanity's Last Exam, ODSA improves accuracy by 4.7 to 18.0 points across three models over a single direct call to the model with no test-time scaling. On LiveCodeBench, where a single direct call already solves most problems, it improves accuracy by 0.6 to 14.0 points.
This study investigates the potential of TimeGPT, a pretrained foundation model for timeseries, for forecasting daily emergency call volumes. Real-world emergency call data, comprising 2,101 daily observations, are used to evaluate both univariate and multivariate forecasting settings under one-step-ahead and seven-day multi-step-ahead scenarios. The multivariate forecasting incorporates temporal, meteorological, demographic, and socioeconomic variables. A chronological training, validation, and test split is adopted to prevent information leakage, while hyperparameters are selected using the validation set through grid search and early stopping when applicable. Forecasting performance is evaluated using multiple error metrics. Its forecasting performance is compared with several established forecasting approaches, including Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors (SARIMAX), Poisson regression (PR), Negative Binomial regression (NBR), Extreme Gradient Boosting (XGBoost), Multilayer Perceptron (MLP), and Long Short-Term Memory (LSTM). The results show that SARIMAX achieves the best overall performance in the univariate setting. In the multivariate setting, NBR and PR provide best performance. These findings indicate that the large-scale pretraining of TimeGPT do not necessarily translate into superior forecasting accuracy for emergency call volumes. Frugal methods trained solely on the available data remain better for this application.
Autonomous microscopy can turn fast acquisition into scientific progress when it decides not only where to measure, but also what information to acquire, at what scan rate, and when to recalibrate the instrument. We develop a budget constrained framework for such multistep structure-property workflows. It combines explicit scientific rewards, action costs, instrument observation models, transfer of pre experiment knowledge into the online campaign, and evaluation kept independent of the controller. The transferred knowledge includes predictor accuracy and computational cost, empirical learner maturity functions that weight model-derived rewards, and drift characteristics established before controller training. Using a real electron energy-loss spectroscopy dataset, we run two complementary campaigns. In the first, the controller chooses between imaging and spectroscopy, image size, and scan speed. In the second, the instrument drifts and the controller also decides when to recalibrate. The campaigns show that response discovery and structure-property learning favor different surrogate models and acquisition strategies. They also show that simple acquisition rules form strong, practical baselines. Independent error analysis reveals where model uncertainty is and is not a reliable learning signal. In the drift campaign, explicit registration greatly improves positional accuracy, and the framework measures what this costs in spectroscopy throughput. Together, these results provide a reproducible way to design, compare, and audit autonomous experiments, and they define what microscopes must provide for deployment: independently executable actions, characterized scan and detector responses, timestamped registration telemetry, and complete action records.
Monocular 3D detection (Mono3D) hinges on recovering depth from a single image, for which the geometric formulation $z \approx f \cdot H / h_{2D}$ is widely adopted due to its simplicity and direct grounding in projective geometry. We show that this formulation harbors a systematic, projection-induced bias: substituting the 2D bounding-box height $h_{2D}$ recovers the depth of an object's visible front surface rather than its center, causing size-dependent underestimation that persists even under ground-truth inputs. We address this front-surface bias with MonoPCA, a unified framework built on a projection-consistent reformulation. At its core, PCDepth retains the measurable $h_{2D}$ and closes the gap to center depth through an orientation- and size-aware correction derived from perspective geometry, eliminating the bias without introducing ill-posed regression targets. Because PCDepth couples depth to predicted object size and orientation, attribute perception becomes essential. We therefore introduce GeoAlign, a teacher-student framework that transfers geometric representations from Vision Foundation Models into a lightweight student, strengthening attribute perception at no inference cost. Across four challenging benchmarks, MonoPCA achieves SoTA performance, including a +1.44% AP$_{3D}$ gain on the KITTI leaderboard and +2.80% on Omni3D$_{OUT}$. Code is available at https://github.com/alanzhangcs/MonoPCA.
Large language models (LLMs) can write the weights of small neural networks directly, without gradient descent. This weight-space programming creates policies for sequential decision-making from a task description alone, but nothing guarantees that a written policy behaves as intended. We combine weight-space programming with probabilistic model checking. Given an environment modeled as a Markov decision process and a requirement in probabilistic computation tree logic (PCTL), an LLM writes all weights of a policy network, optionally repairs them on a dataset of optimal actions, and receives model-checking feedback until the requirement holds or a round limit is reached. The model checker Storm computes, rather than estimates from simulations, the probability that the policy satisfies the requirement, so every returned policy is verified. With two LLMs on three environments, the created policy of each of the 18 runs attains the optimal probability, before any dataset or model-checking feedback. Four of six policies programmed for two transporter jobs also complete three jobs with probability 1, so they generalize to a task that the LLM never saw. The pipeline also programs a policy that satisfies a complex safety requirement expressed with the PCTL until operator in all six runs. In one of them, dataset repair raised the accuracy to 98.4\% but broke the requirement, and one round of model-checking feedback restored it.
Controlled acquisition setups are widely used for high-accuracy multi-view 3D surface reconstruction, as they allow sensor configuration, image resolution, viewpoint distribution, and illumination to be specified. Yet benchmarking remains difficult, since reconstruction accuracy depends jointly on the reconstruction paradigm, object properties, and acquisition regime. Existing benchmarks often restrict viewpoint coverage, resolution, or illumination, which can bias comparisons towards methods suited to the chosen setting. Others lack accurate 3D ground truth or quantified image-to-geometry alignment, limiting their use for metric surface evaluation. We introduce Martine, a benchmark for controlled 3D object surface reconstruction. Martine comprises 28 objects, including transparent materials and fine structures, captured at 9568 x 6376 pixels from 84 views over a spherical cap under both single-light and multi-light regimes. It provides object-independent camera calibration, accurate 3D ground-truth scans, calibration and image-to-geometry accuracy quantified on test images, and 3D property masks for analysing errors by material properties, visibility, curvature and fine structures. Experiments across photogrammetry, classical multi-view stereo, neural implicit surfaces, Gaussian-splatting-based reconstruction, and multi-view photometric stereo show that Martine enables regime-aware evaluation with explicit control over reference-data accuracy. The benchmark resolves sub-millimetric reconstruction differences, with a best observed Chamfer distance of 0.058 mm over the evaluated visible surface.
Knowledge distillation trains a student language model to match a teacher's predictions on the training text. We ask whether learning the teacher's behavior beyond the training text can further improve the student. We investigate this using three input perturbations: additive token noise, blending two documents, and blending neighboring tokens. Specifically, for an early part of pretraining we perturb the inputs and train the student to match the teacher on them, and we spend the rest of the same token budget on ordinary distillation. When pretraining from scratch, the best method improves on standard distillation at both scales we tried, by 0.025 nats at 135M parameters and by 0.020 nats at 0.5B (mean of three seeds). At 135M this is 1.45 times the gain that distillation itself provides over next-token training. We also observe promising downstream transfer: at 0.5B, the best variant improves average zero-shot accuracy by about one percentage point across seven tasks. It also retains its validation-loss advantage after low-bit weight quantization. We also test a data-limited setting in which training data must be reused, and find that the gains persist. Finally, ablations suggest that, for additive noise, perturbing the teacher's inputs contributes most of the gain, while perturbing only the student's inputs yields much less.
Recent work advancing machine translation quality using LLMs for very low resource languages has relied on the injection of lexicon-based prompt components to expose vocabulary to models. However, these approaches run into a ceiling when models must guess sentence meaning purely from plausible combinations of possible word senses mapped to strings. In this paper we test UD-PrMPT, an approach using syntactic analyses in the Universal Dependencies formalism as prompt components to improve MT in five diverse low resource languages: Beja, Chintang, Chukchi, Coptic, and Xibe. Our results show additive improvements which stack on top of the lexicon based approach which are stronger for higher quality manual analyses but consistent for predicted parses as well.
Recent progress on contrastive audio-language models can be attributed mainly to the scale and quality of their training data. Larger training datasets expose models to a broader range of sounds and corresponding audio captions. To exploit this, we collect 1.48 billion audio samples from academic datasets, internet datasets, web collections, Common Crawl, and YouTube, and process them with a multi-stage curation pipeline. Furthermore, we generate short and long captions for samples in our corpus that lack them. We use our dataset to study the scaling behavior of contrastive audio-language models based on the NaFlex architecture for the audio encoder. Our experiments identify trade-offs between model size and training data for a given compute budget. Moreover, our resulting models, CLAP2, outperform prior audio-text models of comparable size on the MAEB and HEAR benchmarks. We release our dataset, captions, model checkpoints, and code.
Ray Tracing enables rendering capabilities that are difficult to achieve with rasterization-based approaches such as 3D Gaussian Splatting (3DGS). However, the training cost of Gaussian-based ray tracing (GRT) is much higher than that of 3DGS. In this paper, we study how ray tracing using Gaussians executes on GPUs and show that a considerable fraction of the training cost arises from how the computation is organized. We develop FastGRT, which decouples ray traversal from blending and reuses work done in the forward pass for the backward pass. FastGRT further incorporates optimizations such as fast gradient accumulation and a hybrid BVH update strategy to tackle other major bottlenecks in the training pipeline. We implement FastGRT in Vulkan. To the best of our knowledge, it is the first Vulkan-based GRT implementation that supports both training and rendering. Compared to 3DGRT, FastGRT improves training time by 1.9$\times$ to 4.97$\times$, while preserving the original rendering formulation. Our code is open source.
Autonomous vehicles use pedestrian crossing intention prediction (PCIP) to participate safely in traffic. Existing models are largely trained under favourable daytime conditions with sufficient daylight. Little is known about the robustness of these models under night-time and low-light conditions, due to the lack of datasets containing night-time footage. This paper investigates how a dataset, named NightSteps, can be constructed and evaluated as a baseline benchmark. The dataset is based on nighttime dashcam footage collected from six different cities around the world, capturing a wide variety of road conditions. The dataset consists of dashcam-recorded point-of-view footage, raw data of pedestrians' locations with corresponding 'track ids' and ground-truth labels containing accurate information about pedestrian crossing behaviour, resulting in a diverse dataset. Gemini, ChatGPT, and human assessment are used to demonstrate the usability of the dataset by predicting crossings in clips from the dataset and comparing the results with the initial ground-truth labels. These results show that human input achieves an accuracy of 88.17% and an F1-score of 80.21%. Predictions from Gemini and ChatGPT perform lower, with accuracies of 60% and 79.13% respectively, and F1-scores of 42.50% and 63.64%. The evaluation demonstrates that the NightSteps dataset can be used for future model evaluation to test the performance of PCIP, with the aim of improving road safety at night.