Large Language Models (LLMs) have demonstrated strong capabilities as autonomous agents through tool use, planning, and decisionmaking abilities, leading to their widespread adoption across diverse tasks.As task complexity grows, multi-agent LLM systems are increasingly used to solve problems collaboratively.However, safety and security of these systems remains largely under-explored.Existing benchmarks and datasets predominantly focus on single-agent settings, failing to capture the unique vulnerabilities of multi-agent dynamics and co-ordination.To address this gap, we introduce Threats and Attacks in Multi-Agent Systems (TAMAS), a benchmark designed to evaluate the robustness and safety of multi-agent LLM systems.TAMAS includes five distinct scenarios comprising 300 adversarial instances across six attack types and 211 tools, along with 100 harmless tasks.We assess system performance across ten backbone LLMs and three agent interaction configurations from Autogen and CrewAI frameworks, highlighting critical challenges and failure modes in current multi-agent deployments.Furthermore, we introduce Effective Robustness Score (ERS) to assess the tradeoff between safety and task effectiveness of these frameworks.Our findings show that multi-agent systems are highly vulnerable to adversarial attacks, underscoring the urgent need for stronger defenses.TAMAS provides a foundation for systematically studying and improving the safety of multi-agent LLM systems.Code and dataset is available at https://github.com/microsoft/TAMAS.

Rohit Sinha, Aditya Sanjiv Kanade, Sai Srinivas Kancheti, Vineeth N. Balasubramanian, Tanuja Ganu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

AI technologies are increasingly deployed in high-stakes domains such as education, healthcare, law, and agriculture to address complex challenges in non-Western contexts. This paper examines eight real-world deployments spanning seven countries and 18 languages, combining 17 interviews with AI developers and domain experts with secondary research. Our findings identify six cross-cutting factors — Language, Institution, Safety, Task, End-User Demography, and Domain — that structured how systems were designed and deployed. These factors were shaped by Sociocultural (diversity, practices), Institutional (resources, policies), and Technological (capabilities, limits) influences. We find that building effective AI systems required extensive collaboration between AI developers and domain experts, with human resources proving more critical to achieving safe and effective outcomes in high-stakes domains than technological expertise alone. Additionally, we present 12 guidelines synthesizing these dynamics for designing AI for social good systems that are culturally grounded, equitable, and responsive to the needs of non-Western contexts.

This study investigates Shiksha Copilot, an AI-assisted lesson planning tool deployed in government schools across Karnataka, India. The system combined LLMs and human expertise through a structured process in which English and Kannada lesson plans were co-created by curators and AI; teachers then further customized these curated plans for their classrooms using their own expertise alongside AI support. Drawing on a large-scale mixed-methods study involving 1,043 teachers and 23 curators, we examine how educators collaborate with AI to generate context-sensitive lesson plans, assess the quality of AI-generated content, and analyze shifts in teaching practices within multilingual, low-resource environments. Our findings show that teachers used Shiksha Copilot both to meet administrative documentation needs and to support their teaching. The tool eased bureaucratic workload, reduced lesson planning time, and lowered teaching-related stress, while promoting a shift toward activity-based pedagogy. However, systemic challenges such as staffing shortages and administrative demands constrained broader pedagogical change. We frame these findings through the lenses of teacher-AI collaboration and communities of practice to examine the effective integration of AI tools in teaching. Finally, we propose design directions for future teacher-centered EdTech, particularly in multilingual and Global South contexts.

Large language models (LLMs) have revolutionized various domains but still struggle with non-Latin scripts and low-resource languages. This paper addresses the critical challenge of improving multilingual performance without extensive fine-tuning. We introduce a novel dynamic learning approach that optimizes prompt strategy, embedding model, and LLM per query at runtime. By adapting configurations dynamically, our method achieves significant improvements over static, best and random baselines. It operates efficiently in both offline and online settings, generalizing seamlessly across new languages and datasets. Leveraging Retrieval-Augmented Generation (RAG) with state-of-the-art multilingual embeddings, we achieve superior task performance across diverse linguistic contexts. Through systematic investigation and evaluation across18 diverse languages using popular question-answering (QA) datasets we show our approach results in 10-15% improvements in multilingual performance over pre-trained models and 4x gains compared to fine-tuned, language-specific models.

Hengyi Wang, Haizhou Shi, Shiwei Tan, Weiyi Qin, Wenyuan Wang, Tunyu Zhang, Akshay Nambi, Tanuja Ganu, Hao Wang. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

A steady increase in the performance of Massively Multilingual Models (MMLMs) has contributed to their rapidly increasing use in data collection pipelines. Interactive Neural Machine Translation (INMT) systems are one class of tools that can utilize MMLMs to promote such data collection in several under-resourced languages. However, these tools are often not adapted to the deployment constraints that native language speakers operate in, as bloated, online inference-oriented MMLMs trained for data-rich languages, drive them. INMT-Lite addresses these challenges through its support of (1) three different modes of Internet-independent deployment and (2) a suite of four assistive interfaces suitable for (3) data-sparse languages. We perform an extensive user study for INMT-Lite with an under-resourced language community, Gondi, to find that INMT-Lite improves the data generation experience of community members along multiple axes, such as cognitive load, task productivity, and interface interaction time and effort, without compromising on the quality of the generated translations.INMT-Lite’s code is open-sourced to further research in this domain.

Spatial representation learning (SRL) aims at learning general-purpose neural network representations from various types of spatial data (e.g., points, polylines, polygons, networks, images, etc.) in their native formats. Learning good spatial representations is a fundamental problem for various downstream applications such as species distribution modeling, weather forecasting, trajectory generation, geographic question answering, etc. Even though SRL has become the foundation of almost all geospatial artificial intelligence (GeoAI) research, we have not yet seen significant efforts to develop an extensive deep learning framework and benchmark to support SRL model development and evaluation. To fill this gap, we propose TorchSpatial, a learning framework and benchmark for location (point) encoding,which is one of the most fundamental data types of spatial representation learning. TorchSpatial contains three key components: 1) a unified location encoding framework that consolidates 15 commonly recognized location encoders, ensuring scalability and reproducibility of the implementations; 2) the LocBench benchmark tasks encompassing 7 geo-aware image classification and 10 geo-aware imageregression datasets; 3) a comprehensive suite of evaluation metrics to quantify geo-aware models’ overall performance as well as their geographic bias, with a novel Geo-Bias Score metric. Finally, we provide a detailed analysis and insights into the model performance and geographic bias of different location encoders. We believe TorchSpatial will foster future advancement of spatial representationlearning and spatial fairness in GeoAI research. The TorchSpatial model framework and LocBench benchmark are available at https://github.com/seai-lab/TorchSpatial, and the Geo-Bias Score evaluation framework is available at https://github.com/seai-lab/PyGBS.

Kabir Ahuja, Harshita Diddee, Rishav Hada, Millicent Ochieng, Krithika Ramesh, Prachi Jain, Akshay Nambi, Tanuja Ganu, Sameer Segal, Mohamed Ahmed, Kalika Bali, Sunayana Sitaram. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023.

Real-time perception requires planned resource utilization. Computational planning in real-time perception is governed by two considerations -- accuracy and latency. There exist run-time decisions (e.g. choice of input resolution) that induce tradeoffs affecting performance on a given hardware, arising from intrinsic (content, e.g. scene clutter) and extrinsic (system, e.g. resource contention) characteristics. Earlier runtime execution frameworks employed rule-based decision algorithms and operated with a fixed algorithm latency budget to balance these concerns, which is sub-optimal and inflexible. We propose Chanakya, a learned approximate execution framework that naturally derives from the streaming perception paradigm, to automatically learn decisions induced by these tradeoffs instead. Chanakya is trained via novel rewards balancing accuracy and latency implicitly, without approximating either objectives. Chanakya simultaneously considers intrinsic and extrinsic context, and predicts decisions in a flexible manner. Chanakya, designed with low overhead in mind, outperforms state-of-the-art static and dynamic execution policies on public datasets on both server GPUs and edge devices.

Pre-trained multilingual language models are gaining popularity due to their cross-lingual zero-shot transfer ability, but these models do not perform equally well in all languages. Evaluating task-specific performance of a model in a large number of languages is often a challenge due to lack of labeled data, as is targeting improvements in low performing languages through few-shot learning. We present a tool - LITMUS Predictor - that can make reliable performance projections for a fine-tuned task-specific model in a set of languages without test and training data, and help strategize data labeling efforts to optimize performance and fairness objectives.

Accurately predicting the energy consumption of an electric vehicle (EV) under real-world circumstances (such as varying road, traffic, weather conditions, etc.) is critical for a number of decisions like range estimation and route planning. A major concern for electric vehicle owners is the uncertain nature of the battery consumption. This results in the “range anxiety” and reluctance from users for mass adoption of EVs, since they are concerned about untimely drainage of battery. Even at the organizational level, a company running a fleet of electric vehicles must understand the battery consumption profiles accurately for tasks such as route and driver planning, battery sizing, maintenance planning, etc.

Seamless access to information in a rapidly globalizing world demands for availability of information across, ideally all but at the least a large number of, languages. Machine translation has been proposed as a technological solution to this complex problem. However, despite seven decades of research, and recently seen rapid progress in the field - thanks to deep learning and availability of large data-sets, perfect machine translation across a large number of the world’s languages still remains elusive. In fact, it is a distant and perhaps even an impossible goal. Erroneous translations, on the other hand, can be detrimental in critical situations such as talking to a law enforcement officer; or, they could potentially perpetuate social biases or stereotypes, for instance, by producing mis-gendered translations. In this work, we argue that language translation is inherently a socio-technical system, which has to be viewed, studied, and optimized for, as such. The need and context of translation, the socio-demographic factors behind the human translators as well as the consumers of the translated content affect the complexity of the translation system, as much as the accuracy of the technology and its interface. Through a series of case studies on mixed-initiative interaction based approach to translation, we bring out the various socio-technical factors and their complex interactions that one has to bear in mind while designing for the ideal human-machine translation systems. Through these observations, we make multiple recommendations which, at the core, suggest that ”solving” translation in the real sense would require more coordinated efforts between the technical (NLP) and social communities (HCI + CSCW + DEV).

This paper presents a deep learning approach for a versatile Micro-climate prediction framework (DeepMC). Micro climate predictions are of critical importance across various applications, such as Agriculture, Forestry, Energy, Search & Rescue, etc. To the best of our knowledge, there is no other single framework which can accurately predict various micro-climate entities using Internet of Things (IoT)data. We present a generic framework (DeepMC) which predicts various climatic parameters such as soil moisture, humidity, windspeed, radiation, temperature based on the requirement over a period of 12 hours - 120 hours with a varying resolution of 1 hour - 6hours, respectively. This framework proposes the following new ideas: 1) Localization of weather forecast to IoT sensors by fusing weather station forecasts with the decomposition of IoT data at multiple scales and 2) A multi-scale encoder and two levels of attention mechanisms which learns a latent representation of the interaction between various resolutions of the IoT sensor data and weather station forecasts. We present multiple real-world agricultural and energy scenarios, and report results with uncertainty estimates from the live deployment of DeepMC, which demonstrate that DeepMC outperforms various baseline methods and reports 90%+ accuracy with tight error bounds.

With the increasing adaption of solar energy worldwide, there is a huge interest to develop systems that help drive efficiency during manufacturing and ongoing operations. Due to various real-world conditions and processes, solar panels develop faults during their manufacturing and operations. The objective of this work is to build an End-to-End Fault Detection system to detect and localize faults in solar panels based on their Electroluminescence (EL) Imaging. Today, the majority of fault detection happens through manual inspection of EL images. To this end, we propose the design and implementation of an end-to-end system that firstly divides the solar panel into individual solar cells and then passes these cell images through a classification + detection pipeline for identifying the fault type and localizing the faults inside a cell. We propose a hybrid architecture that contains an ensemble of multiple CNN model architectures for classification and detection. The ensemble is capable of serving both - monocrystalline and polycrystalline solar panels. The proposed system significantly helps in increasing the efficiency of solar panels and reducing warranty and repair costs. We demonstrate the performance of the proposed system using an open EL image dataset with 95% of cell-level fault prediction accuracy and high recall. The proposed algorithms are applicable and can be extended for other solar applications that use RGB, EL, or thermal imaging techniques.

A significant amount of energy is wasted by electrical appliances when they operate inefficiently either due to anomalies and/or incorrect usage. To address this problem, we present SocketWatch - an autonomous appliance monitoring system. SocketWatch is positioned between a wall socket and an appliance. SocketWatch learns the behavioral model of the appliance by analyzing its active and reactive power consumption patterns. It detects appliance malfunctions by observing any marked deviations from these patterns. SocketWatch is inexpensive and is easy to use: it neither requires any enhancement to the appliances nor to the power sockets nor any communication infrastructure. Moreover, the decentralized approach avoids communication latency and costs, and preserves data privacy. Real world experiments with multiple appliances indicate that SocketWatch can be an effective and inexpensive solution for reducing electricity wastage.

Many electricity suppliers around the world are deploying smart meters to gather fine-grained spatiotemporal consumption data and to effectively manage the collective demand of their consumer base. In this paper, we introduce a structured framework and a discriminative index that can be used to segment the consumption data along multiple contextual dimensions such as locations, communities, seasons, weather patterns, holidays, etc. The generated segments can enable various higher-level applications such as usage-specific tariff structures, theft detection, consumer-specific demand response programs, etc. Our framework is also able to track consumers’ behavioral changes, evaluate different temporal aggregations, and identify main characteristics which define a cluster.

We address the problems of sparse multiclass and multi-label classifier design and devise new algorithms using margin based ideas. Many online applications such as image classification or text categorization demand fast inference. State-of-the-art classifiers such as Support Vector Machines (SVM) are not preferred in such applications because of slow inference, which is mainly due to the large number of support vectors required to form the SVM classifier. We propose algorithms which solve primal problems directly by greedily adding the required number of basis functions into the classifier model. Experiments on various real-world data sets demonstrate that the proposed algorithms output significantly smaller number of basis functions, while achieving nearly the same generalization performance as that given by SVM and other state-of-the-art sparse classifiers. This enables the classifiers to perform faster inference, thereby making the proposed algorithms powerful alternatives to existing approaches.

The Indian electricity sector, despite having the world's fifth largest installed capacity, suffers from a 12.9% peaking shortage. This shortage could be alleviated, if a large number of deferrable loads, particularly the high powered ones, could be moved from on-peak to off-peak times. However, conventional Demand Side Management (DSM) strategies may not be suitable for India as the local conditions usually favor inexpensive solutions with minimal dependence on the pre-existing infrastructure. In this work, we present a completely autonomous DSM controller called the nPlug. nPlug is positioned between the wall socket and deferrable load(s) such as water heaters, washing machines, and electric vehicles. nPlugs combine local sensing and analytics to infer peak periods as well as supply-demand imbalance conditions. They schedule attached appliances in a decentralized manner to alleviate peaks whenever possible without violating the requirements of consumers. nPlugs do not require any manual intervention by the end consumer nor any communication infrastructure nor any enhancements to the appliances or the power grids. Some of nPlug's capabilities are demonstrated using experiments on a combination of synthetic and real data collected from plug-level energy monitors. Our results indicate that nPlug can be an effective and inexpensive technology to address the peaking shortage. This technology could potentially be integrated into millions of future deferrable loads: appliances, electric vehicle (EV) chargers, heat pumps, water heaters, etc.