While multilingual large language models (LLMs) perform well on high-level tasks like translation and question answering, their ability to handle grammatical gender and morphological agreement remains underexplored.In morphologically rich languages, gender influences verb conjugation, pronouns, and even first-person constructions with explicit and implicit mentions to gender.We thus introduce MORPHOGEN a morphologically grounded largescale benchmark dataset for evaluating genderaware generation in three typologically diverse grammatically gendered languages i.e.French, Arabic and Hindi.The core task, GENFORM, requires models to rewrite a first-person sentence in the opposite gender while preserving its meaning and structure.We construct a highquality synthetic dataset spanning French, Arabic, and Hindi, and benchmark 15 popular multilingual LLMs (2B-70B) on their ability to perform this transformation.Our results reveal gaps and interesting insights into the handling of morphological gender in current models.MORPHOGEN offers a focused diagnostic lens for gender-aware language modeling and lays the groundwork for future research on inclusive and morphology-sensitive NLP.

In open-ended generative tasks like narrative writing or dialogue, large language models often exhibit cultural biases, showing limited knowledge and generating templated outputs for less prevalent cultures. Recent works show that these biases may stem from uneven cultural representation in pretraining corpora. This work investigates how pretraining leads to biased culture-conditioned generations by analyzing how models associate entities with cultures based on pretraining data patterns. We propose the MEMOED framework (MEMOrization from prEtraining Document) to determine whether a generation for a culture arises from memorization. Using MEMOED on culture-conditioned generations about food and clothing for 110 cultures, we find that high-frequency cultures in pretraining data yield more generations with memorized symbols, while some low-frequency cultures produce none. Additionally, the model favors generating entities with extraordinarily high frequency regardless of the conditioned culture, reflecting biases toward frequent pretraining terms irrespective of relevance. We hope that the MEMOED framework and our insights will inspire more works on attributing model performance on pretraining data.

Canopy height is a fundamental metric for extracting valuable information about forested areas. Over the past decade, the light detection and ranging (LiDAR) technology has provided a straightforward method for measuring canopy height using various platforms, including terrestrial, uncrewed aerial vehicles (UAVs), airborne, and satellite sensors. However, despite its global reach, spaceborne LiDAR data suffers from a sparse sampling pattern that fails to provide continuous global coverage. In contrast, satellites like LANDSAT deliver seamless and extensive coverage of the Earth’s surface through spectral data. This study aims to develop a deep learning model to infer canopy heights from sparsely observed LiDAR data, utilizing the multisensor spectral data from spaceborne platforms. Specifically tailored for localized sites, the model focuses on regional-level canopy height estimation by leveraging the relationship between canopy height and multisensor time-series data from Landsat, Sentinel-2, and Sentinel-1. We first demonstrate the importance of integrating multisensor data by training three separate models: one using only Landsat data, one using only Sentinel-2 data, and a multimodal model that incorporates Landsat, Sentinel 1, and Sentinel 2 data to estimate LiDAR-derived canopy height. These models were tested on two sites in Indiana—Tippecanoe and Monroe counties—where the multimodal approach produced the best results, achieving RMSEs of 3.895 and 4.993 m, respectively. We then tested our multimodal model in two additional counties—Baker County, FL, USA and Piute County, UT, USA—where the model achieved an RMSE of 5.397 and 3.742 m, respectively.

Although the rise of Large Language Models (LLMs) in enterprise settings brings new opportunities and capabilities, it also brings challenges, such as the risk of generating inappropriate, biased, or misleading content that violates regulations and can have legal concerns. To alleviate this, we present "LLMGuard", a tool that monitors user interactions with an LLM application and flags content against specific behaviours or conversation topics. To do this robustly, LLMGuard employs an ensemble of detectors.

Small-World phenomena is a well studied area in complex networks. In traditional networks, the application of small world is to reduce the path length between any two nodes participating in the communication. However, in Wireless Sensor networks (WSNs), less work has been done to address small world problems, as the communication pattern is different. The introduction of small world characteristics in WSN results in transmitting information from one node to another in fewer numbers of hops than in conventional WSN. In this paper the sensor network is modeled as a small world network by introducing Directional Sensor Nodes (DSNs) and hence to find a route with minimum path length. The performance of the proposed approach was done using extensive simulation. The results indicate that the proposed approach, can find a path with minimum hops from source to sink which can route the data in minimum time and hence make the network more energy efficient.