We introduce SigmaCollab, a dataset enabling research on physically situated human-AI collaboration. The dataset consists of a set of 85 sessions in which untrained participants were guided by a mixed-reality assistive AI agent in performing procedural tasks in the physical world. SigmaCollab includes a set of rich, multimodal data streams, such as the participant and system audio, egocentric camera views from the head-mounted device, depth maps, head, hand and gaze tracking information, as well as additional annotations performed post-hoc. While the dataset is relatively small in size (~ 14 hours), its application-driven and interactive nature brings to the fore novel research challenges for human-AI collaboration, and provides more realistic testing grounds for various AI models operating in this space. In future work, we plan to use the dataset to construct a set of benchmarks for physically situated collaboration in mixed-reality task assistive scenarios. SigmaCollab is available at https://github.com/microsoft/SigmaCollab.

Recent advances in multi-modal models have demonstrated strong performance in tasks such as image generation and reasoning. However, applying these models to the fire domain remains challenging due to the lack of publicly available datasets with high-quality fire domain annotations. To address this gap, we introduce DetectiumFire, a large-scale, multi-modal dataset comprising of 22.5k high-resolution fire-related images and 2.5k real-world fire-related videos covering a wide range of fire types, environments, and risk levels. The data are annotated with both traditional computer vision labels (e.g., bounding boxes) and detailed textual prompts describing the scene, enabling applications such as synthetic data generation and fire risk reasoning. DetectiumFire offers clear advantages over existing benchmarks in scale, diversity, and data quality, significantly reducing redundancy and enhancing coverage of real-world scenarios. We validate the utility of DetectiumFire across multiple tasks, including object detection, diffusion-based image generation, and vision-language reasoning. Our results highlight the potential of this dataset to advance fire-related research and support the development of intelligent safety systems. We release DetectiumFire to promote broader exploration of fire understanding in the AI community. The dataset is available at https://kaggle.com/datasets/38b79c344bdfc55d1eed3d22fbaa9c31fad45e27edbbe9e3c529d6e5c4f93890

Accurately forecasting central bank policy decisions, particularly those of the Federal Open Market Committee (FOMC) has become increasingly important amid heightened economic uncertainty. While prior studies have used monetary policy texts to predict rate changes, most rely on static classification models that overlook the deliberative nature of policymaking. This study proposes a novel framework that structurally imitates the FOMC's collective decision-making process by modeling multiple large language models (LLMs) as interacting agents. Each agent begins with a distinct initial belief and produces a prediction based on both qualitative policy texts and quantitative macroeconomic indicators. Through iterative rounds, agents revise their predictions by observing the outputs of others, simulating deliberation and consensus formation. To enhance interpretability, we introduce a latent variable representing each agent's underlying belief (e.g., hawkish or dovish), and we theoretically demonstrate how this belief mediates the perception of input information and interaction dynamics. Empirical results show that this debate-based approach significantly outperforms standard LLMs-based baselines in prediction accuracy. Furthermore, the explicit modeling of beliefs provides insights into how individual perspectives and social influence shape collective policy forecasts.

We evaluate whether persona-based prompting improves Large Language Model (LLM) performance on macroeconomic forecasting tasks. Using 2,368 economics-related personas from the PersonaHub corpus, we prompt GPT-4o to replicate the ECB Survey of Professional Forecasters across 50 quarterly rounds (2013-2025). We compare the persona-prompted forecasts against the human experts panel, across four target variables (HICP, core HICP, GDP growth, unemployment) and four forecast horizons. We also compare the results against 100 baseline forecasts without persona descriptions to isolate its effect. We report two main findings. Firstly, GPT-4o and human forecasters achieve remarkably similar accuracy levels, with differences that are statistically significant yet practically modest. Our out-of-sample evaluation on 2024-2025 data demonstrates that GPT-4o can maintain competitive forecasting performance on unseen events, though with notable differences compared to the in-sample period. Secondly, our ablation experiment reveals no measurable forecasting advantage from persona descriptions, suggesting these prompt components can be omitted to reduce computational costs without sacrificing accuracy. Our results provide evidence that GPT-4o can achieve competitive forecasting accuracy even on out-of-sample macroeconomic events, if provided with relevant context data, while revealing that diverse prompts produce remarkably homogeneous forecasts compared to human panels.

Machine learning methods are increasingly applied in medical imaging, yet many reported improvements lack statistical robustness: recent works have highlighted that small but significant performance gains are highly likely to be false positives. However, these analyses do not take \emph{underspecification} into account -- the fact that models achieving similar validation scores may behave differently on unseen data due to random initialization or training dynamics. Here, we extend a recent statistical framework modeling false outperformance claims to include underspecification as an additional variance component. Our simulations demonstrate that even modest seed variability ($\sim1\%$) substantially increases the evidence required to support superiority claims. Our findings underscore the need for explicit modeling of training variance when validating medical imaging systems.

Machine learning models are integrated into many real-world applications. Since these models contain artifacts of potentially sensitive training data, this raises concerns about data confidentiality and user privacy. The ability to remove specific training data from a model has emerged as a key mechanism to enforce, for instance, the "right to be forgotten" promoted by the European GDPR law [17] or the "right to erase" in the Canadian CPPA legislation [29]. Approximate machine unlearning aims to find efficient mechanisms, avoiding the cost of learning a new model from scratch over the training dataset deprived of the sensitive data. Privacy is not the only application of machine unlearning. It has been proven useful as a defense against backdoor attacks by annihilating the influence of the poisoned training data [45, 38], or to improve fairness by removing data that induce biases in the training set. Another scenario is the derivation of a restricted public model from a powerful private model learned on some sensitive data [12]. The accuracy of the public model should be on par with or slightly degraded compared to the private model.

Recent advancements in large language models (LLMs) have enabled significant progress in decision-making and task planning for embodied autonomous agents. However, most existing methods still struggle with complex, long-horizon tasks because they rely on a monolithic trajectory that entangles all past decisions and observations, attempting to solve the entire task in a single unified process. To address this limitation, we propose ReAcTree, a hierarchical task-planning method that decomposes a complex goal into more manageable subgoals within a dynamically constructed agent tree. Each subgoal is handled by an LLM agent node capable of reasoning, acting, and further expanding the tree, while control flow nodes coordinate the execution strategies of agent nodes. In addition, we integrate two complementary memory systems: each agent node retrieves goal-specific, subgoal-level examples from episodic memory and shares environment-specific observations through working memory. Experiments on the WAH-NL and ALFRED datasets demonstrate that ReAcTree consistently outperforms strong task-planning baselines such as ReAct across diverse LLMs. Notably, on WAH-NL, ReAcTree achieves a 61% goal success rate with Qwen 2.5 72B, nearly doubling ReAct's 31%.

Precise semantic segmentation of crops and weeds is necessary for agricultural weeding robots. However, training deep learning models requires large annotated datasets, which are costly to obtain in real fields. Synthetic data can reduce this burden, but the gap between simulated and real images remains a challenge. In this paper, we present a pipeline for procedural generation of synthetic crop-weed images using Blender, producing annotated datasets under diverse conditions of plant growth, weed density, lighting, and camera angle. We benchmark several state-of-the-art segmentation models on synthetic and real datasets and analyze their cross-domain generalization. Our results show that training on synthetic images leads to a sim-to-real gap of 10%, surpassing previous state-of-the-art methods. Moreover, synthetic data demonstrates good generalization properties, outperforming real datasets in cross-domain scenarios. These findings highlight the potential of synthetic agricultural datasets and support hybrid strategies for more efficient model training.

Cats and humans differ in ocular anatomy. Most notably, Felis Catus (domestic cats) have vertically elongated pupils linked to ambush predation; yet, how such specializations manifest in downstream visual representations remains incompletely understood. We present a unified, frozen-encoder benchmark that quantifies feline-human cross-species representational alignment in the wild, across convolutional networks, supervised Vision Transformers, windowed transformers, and self-supervised ViTs (DINO), using layer-wise Centered Kernel Alignment (linear and RBF) and Representational Similarity Analysis, with additional distributional and stability tests reported in the paper. Across models, DINO ViT-B/16 attains the most substantial alignment (mean CKA-RBF $\approx0.814$, mean CKA-linear $\approx0.745$, mean RSA $\approx0.698$), peaking at early blocks, indicating that token-level self-supervision induces early-stage features that bridge species-specific statistics. Supervised ViTs are competitive on CKA yet show weaker geometric correspondence than DINO (e.g., ViT-B/16 RSA $\approx0.53$ at block8; ViT-L/16 $\approx0.47$ at block14), revealing depth-dependent divergences between similarity and representational geometry. CNNs remain strong baselines but below plain ViTs on alignment, and windowed transformers underperform plain ViTs, implicating architectural inductive biases in cross-species alignment. Results indicate that self-supervision coupled with ViT inductive biases yields representational geometries that more closely align feline and human visual systems than widely used CNNs and windowed Transformers, providing testable neuroscientific hypotheses about where and how cross-species visual computations converge. We release our code and dataset for reference and reproducibility.

The development of clinically reliable artificial intelligence (AI) systems for mammography is hindered by profound heterogeneity in data quality, metadata standards, and population distributions across public datasets. This heterogeneity introduces dataset-specific biases that severely compromise the generalizability of the model, a fundamental barrier to clinical deployment. We present MammoClean, a public framework for standardization and bias quantification in mammography datasets. MammoClean standardizes case selection, image processing (including laterality and intensity correction), and unifies metadata into a consistent multi-view structure. We provide a comprehensive review of breast anatomy, imaging characteristics, and public mammography datasets to systematically identify key sources of bias. Applying MammoClean to three heterogeneous datasets (CBIS-DDSM, TOMPEI-CMMD, VinDr-Mammo), we quantify substantial distributional shifts in breast density and abnormality prevalence. Critically, we demonstrate the direct impact of data corruption: AI models trained on corrupted datasets exhibit significant performance degradation compared to their curated counterparts. By using MammoClean to identify and mitigate bias sources, researchers can construct unified multi-dataset training corpora that enable development of robust models with superior cross-domain generalization. MammoClean provides an essential, reproducible pipeline for bias-aware AI development in mammography, facilitating fairer comparisons and advancing the creation of safe, effective systems that perform equitably across diverse patient populations and clinical settings. The open-source code is publicly available from: https://github.com/Minds-R-Lab/MammoClean.