Aditya Gupta

Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao et al. · allen-ai, amazon-science

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting.

Pei Zhou, Aman Madaan, Srividya Pranavi Potharaju et al. · cmu

"Thinking is for Doing." Humans can infer other people's mental states from observations--an ability called Theory-of-Mind (ToM)--and subsequently act pragmatically on those inferences. Existing question answering benchmarks such as ToMi ask models questions to make inferences about beliefs of characters in a story, but do not test whether models can then use these inferences to guide their actions. We propose a new evaluation paradigm for large language models (LLMs): Thinking for Doing (T4D), which requires models to connect inferences about others' mental states to actions in social scenarios. Experiments on T4D demonstrate that LLMs such as GPT-4 and PaLM 2 seemingly excel at tracking characters' beliefs in stories, but they struggle to translate this capability into strategic action. Our analysis reveals the core challenge for LLMs lies in identifying the implicit inferences about mental states without being explicitly asked about as in ToMi, that lead to choosing the correct action in T4D. To bridge this gap, we introduce a zero-shot prompting framework, Foresee and Reflect (FaR), which provides a reasoning structure that encourages LLMs to anticipate future challenges and reason about potential actions. FaR boosts GPT-4's performance from 50% to 71% on T4D, outperforming other prompting methods such as Chain-of-Thought and Self-Ask. Moreover, FaR generalizes to diverse out-of-distribution story structures and scenarios that also require ToM inferences to choose an action, consistently outperforming other methods including few-shot in-context learning.

Pranjal Aggarwal, Aman Madaan, Ankit Anand et al. · cmu

Large language models (LLMs) are now available from cloud API providers in various sizes and configurations. While this diversity offers a broad spectrum of choices, effectively leveraging the options to optimize computational cost and performance remains challenging. In this work, we present Automix, an approach that strategically routes queries to larger LMs, based on the approximate correctness of outputs from a smaller LM. Central to Automix are two key technical contributions. First, it has a few-shot self-verification mechanism, which estimates the reliability of its own outputs without requiring extensive training. Second, given that self-verification can be noisy, it employs a POMDP based router that can effectively select an appropriately sized model, based on answer confidence. Experiments across five language models and five challenging datasets show that Automix consistently surpasses strong baselines, reducing computational cost by over 50% for comparable performance.

Jingfeng Yang, Aditya Gupta, Shyam Upadhyay et al. · amazon-science

Understanding tables is an important aspect of natural language understanding. Existing models for table understanding require linearization of the table structure, where row or column order is encoded as an unwanted bias. Such spurious biases make the model vulnerable to row and column order perturbations. Additionally, prior work has not thoroughly modeled the table structures or table-text alignments, hindering the table-text understanding ability. In this work, we propose a robust and structurally aware table-text encoding architecture TableFormer, where tabular structural biases are incorporated completely through learnable attention biases. TableFormer is (1) strictly invariant to row and column orders, and, (2) could understand tables better due to its tabular inductive biases. Our evaluations showed that TableFormer outperforms strong baselines in all settings on SQA, WTQ and TabFact table reasoning datasets, and achieves state-of-the-art performance on SQA, especially when facing answer-invariant row and column order perturbations (6% improvement over the best baseline), because previous SOTA models' performance drops by 4% - 6% when facing such perturbations while TableFormer is not affected.

Rahul Goel, Waleed Ammar, Aditya Gupta et al.

Research interest in task-oriented dialogs has increased as systems such as Google Assistant, Alexa and Siri have become ubiquitous in everyday life. However, the impact of academic research in this area has been limited by the lack of datasets that realistically capture the wide array of user pain points. To enable research on some of the more challenging aspects of parsing realistic conversations, we introduce PRESTO, a public dataset of over 550K contextual multilingual conversations between humans and virtual assistants. PRESTO contains a diverse array of challenges that occur in real-world NLU tasks such as disfluencies, code-switching, and revisions. It is the only large scale human generated conversational parsing dataset that provides structured context such as a user's contacts and lists for each example. Our mT5 model based baselines demonstrate that the conversational phenomenon present in PRESTO are challenging to model, which is further pronounced in a low-resource setup.

Aditya Gupta, Shiva Maharaj, Nicholas Polson et al.

We propose a neural network-based approach to calculate the value of a chess square-piece combination. Our model takes a triplet (Color, Piece, Square) as an input and calculates a value that measures the advantage/disadvantage of having this piece on this square. Our methods build on recent advances in chess AI, and can accurately assess the worth of positions in a game of chess. The conventional approach assigns fixed values to pieces $(\symking=\infty, \symqueen=9, \symrook=5, \symbishop=3, \symknight=3, \sympawn=1)$. We enhance this analysis by introducing marginal valuations. We use deep Q-learning to estimate the parameters of our model. We demonstrate our method by examining the positioning of Knights and Bishops, and also provide valuable insights into the valuation of pawns. Finally, we conclude by suggesting potential avenues for future research.

Ayush Kaushal, Aditya Gupta, Shyam Upadhyay et al.

A naive application of state-of-the-art bidirectional encoders for streaming sequence tagging would require encoding each token from scratch for each new token in an incremental streaming input (like transcribed speech). The lack of re-usability of previous computation leads to a higher number of Floating Point Operations (or FLOPs) and higher number of unnecessary label flips. Increased FLOPs consequently lead to higher wall-clock time and increased label flipping leads to poorer streaming performance. In this work, we present a Hybrid Encoder with Adaptive Restart (HEAR) that addresses these issues while maintaining the performance of bidirectional encoders over the offline (or complete) inputs while improving performance on streaming (or incomplete) inputs. HEAR has a Hybrid unidirectional-bidirectional encoder architecture to perform sequence tagging, along with an Adaptive Restart Module (ARM) to selectively guide the restart of bidirectional portion of the encoder. Across four sequence tagging tasks, HEAR offers FLOP savings in streaming settings upto 71.1% and also outperforms bidirectional encoders for streaming predictions by upto +10% streaming exact match.

Aditya Gupta, Asheesh Singh

Agriculture, as the cornerstone of human civilization, constantly seeks to integrate technology for enhanced productivity and sustainability. This paper introduces $\textit{Agri-GNN}$, a novel Genotypic-Topological Graph Neural Network Framework tailored to capture the intricate spatial and genotypic interactions of crops, paving the way for optimized predictions of harvest yields. $\textit{Agri-GNN}$ constructs a Graph $\mathcal{G}$ that considers farming plots as nodes, and then methodically constructs edges between nodes based on spatial and genotypic similarity, allowing for the aggregation of node information through a genotypic-topological filter. Graph Neural Networks (GNN), by design, consider the relationships between data points, enabling them to efficiently model the interconnected agricultural ecosystem. By harnessing the power of GNNs, $\textit{Agri-GNN}$ encapsulates both local and global information from plants, considering their inherent connections based on spatial proximity and shared genotypes, allowing stronger predictions to be made than traditional Machine Learning architectures. $\textit{Agri-GNN}$ is built from the GraphSAGE architecture, because of its optimal calibration with large graphs, like those of farming plots and breeding experiments. $\textit{Agri-GNN}$ experiments, conducted on a comprehensive dataset of vegetation indices, time, genotype information, and location data, demonstrate that $\textit{Agri-GNN}$ achieves an $R^2 = .876$ in yield predictions for farming fields in Iowa. The results show significant improvement over the baselines and other work in the field. $\textit{Agri-GNN}$ represents a blueprint for using advanced graph-based neural architectures to predict crop yield, providing significant improvements over baselines in the field.

Geunseob Oh, Rahul Goel, Chris Hidey et al.

Semantic parsing (SP) is a core component of modern virtual assistants like Google Assistant and Amazon Alexa. While sequence-to-sequence-based auto-regressive (AR) approaches are common for conversational semantic parsing, recent studies employ non-autoregressive (NAR) decoders and reduce inference latency while maintaining competitive parsing quality. However, a major drawback of NAR decoders is the difficulty of generating top-k (i.e., k-best) outputs with approaches such as beam search. To address this challenge, we propose a novel NAR semantic parser that introduces intent conditioning on the decoder. Inspired by the traditional intent and slot tagging parsers, we decouple the top-level intent prediction from the rest of a parse. As the top-level intent largely governs the syntax and semantics of a parse, the intent conditioning allows the model to better control beam search and improves the quality and diversity of top-k outputs. We introduce a hybrid teacher-forcing approach to avoid training and inference mismatch. We evaluate the proposed NAR on conversational SP datasets, TOP & TOPv2. Like the existing NAR models, we maintain the O(1) decoding time complexity while generating more diverse outputs and improving the top-3 exact match (EM) by 2.4 points. In comparison with AR models, our model speeds up beam search inference by 6.7 times on CPU with competitive top-k EM.

Gheorghe Comanici, Eric Bieber, Mike Schaekermann et al. · amazon-science, baidu

In this report, we introduce the Gemini 2.X model family: Gemini 2.5 Pro and Gemini 2.5 Flash, as well as our earlier Gemini 2.0 Flash and Flash-Lite models. Gemini 2.5 Pro is our most capable model yet, achieving SoTA performance on frontier coding and reasoning benchmarks. In addition to its incredible coding and reasoning skills, Gemini 2.5 Pro is a thinking model that excels at multimodal understanding and it is now able to process up to 3 hours of video content. Its unique combination of long context, multimodal and reasoning capabilities can be combined to unlock new agentic workflows. Gemini 2.5 Flash provides excellent reasoning abilities at a fraction of the compute and latency requirements and Gemini 2.0 Flash and Flash-Lite provide high performance at low latency and cost. Taken together, the Gemini 2.X model generation spans the full Pareto frontier of model capability vs cost, allowing users to explore the boundaries of what is possible with complex agentic problem solving.

Sander Riisøen Jyhne, Aditya Gupta, Ben Worsley et al.

We present NordFKB, a fine-grained benchmark dataset for geospatial AI in Norway, derived from the authoritative, highly accurate, national Felles KartdataBase (FKB). The dataset contains high-resolution orthophotos paired with detailed annotations for 36 semantic classes, including both per-class binary segmentation masks in GeoTIFF format and COCO-style bounding box annotations. Data is collected from seven geographically diverse areas, ensuring variation in climate, topography, and urbanization. Only tiles containing at least one annotated object are included, and training/validation splits are created through random sampling across areas to ensure representative class and context distributions. Human expert review and quality control ensures high annotation accuracy. Alongside the dataset, we release a benchmarking repository with standardized evaluation protocols and tools for semantic segmentation and object detection, enabling reproducible and comparable research. NordFKB provides a robust foundation for advancing AI methods in mapping, land administration, and spatial planning, and paves the way for future expansions in coverage, temporal scope, and data modalities.

Lianhui Qin, Aditya Gupta, Shyam Upadhyay et al.

Everyday conversations require understanding everyday events, which in turn, requires understanding temporal commonsense concepts interwoven with those events. Despite recent progress with massive pre-trained language models (LMs) such as T5 and GPT-3, their capability of temporal reasoning in dialogs remains largely under-explored. In this paper, we present the first study to investigate pre-trained LMs for their temporal reasoning capabilities in dialogs by introducing a new task and a crowd-sourced English challenge set, TIMEDIAL. We formulate TIME-DIAL as a multiple-choice cloze task with over 1.1K carefully curated dialogs. Empirical results demonstrate that even the best performing models struggle on this task compared to humans, with 23 absolute points of gap in accuracy. Furthermore, our analysis reveals that the models fail to reason about dialog context correctly; instead, they rely on shallow cues based on existing temporal patterns in context, motivating future research for modeling temporal concepts in text and robust contextual reasoning about them. The dataset is publicly available at: https://github.com/google-research-datasets/timedial.

Aditya Gupta, Jiacheng Xu, Shyam Upadhyay et al.

Disfluencies is an under-studied topic in NLP, even though it is ubiquitous in human conversation. This is largely due to the lack of datasets containing disfluencies. In this paper, we present a new challenge question answering dataset, Disfl-QA, a derivative of SQuAD, where humans introduce contextual disfluencies in previously fluent questions. Disfl-QA contains a variety of challenging disfluencies that require a more comprehensive understanding of the text than what was necessary in prior datasets. Experiments show that the performance of existing state-of-the-art question answering models degrades significantly when tested on Disfl-QA in a zero-shot setting.We show data augmentation methods partially recover the loss in performance and also demonstrate the efficacy of using gold data for fine-tuning. We argue that we need large-scale disfluency datasets in order for NLP models to be robust to them. The dataset is publicly available at: https://github.com/google-research-datasets/disfl-qa.

Aditya Gupta, Abhishek Shamra

This paper addresses the design and development of an autonomous biped robot using master and worker combination of controllers. In addition, the bot is wirelessly controllable. The work presented here explains the walking pattern, system control and actuator control techniques for 10 Degree of Freedom (DOF) biped humanoid. Bi-pedal robots have better mobility than conventional wheeled robots, but they tend to topple easily. In order to walk stably in various environments, such as on rough terrain, up and down slopes, or in regions containing obstacles, it is necessary, that robot should adapt to the ground conditions with a foot motion, as well as maintain its stability with a torso motion. It is desirable to select a walking pattern that requires small torque and velocity of the joint actuators. The work proposed a low cost solution using open source hardware-software and application. The work extends to develop and implement new algorithms by adding gyroscope and accelerometer to further the research in the field of biped robots.

Aditya Gupta, Samarth Chitgopekar, Alexander Kim et al.

We present our work on electric bicycles (e-bikes) and their implications for policymakers in the United States. E-bikes have gained significant popularity as a fast and eco-friendly transportation option. As we strive for a sustainable energy plan, understanding the growth and impact of e-bikes is crucial for policymakers. Our mathematical modeling offers insights into the value of e-bikes and their role in the future. Using an ARIMA model, a supervised machine-learning algorithm, we predicted the growth of e-bike sales in the U.S. Our model, trained on historical sales data from January 2006 to December 2022, projected sales of 1.3 million units in 2025 and 2.113 million units in 2028. To assess the factors contributing to e-bike usage, we employed a Random Forest regression model. The most significant factors influencing e-bike sales growth were disposable personal income and popularity. Furthermore, we examined the environmental and health impacts of e-bikes. Through Monte Carlo simulations, we estimated the reduction in carbon emissions due to e-bike use and the calories burned through e-biking. Our findings revealed that e-bike usage in the U.S. resulted in a reduction of 15,737.82 kilograms of CO2 emissions in 2022. Additionally, e-bike users burned approximately 716,630.727 kilocalories through their activities in the same year. Our research provides valuable insights for policymakers, emphasizing the potential of e-bikes as a sustainable transportation solution. By understanding the growth factors and quantifying the environmental and health benefits, policymakers can make informed decisions about integrating e-bikes into future energy and transportation strategies.

Zixuan Pan, Zihao Wei, Yidong Huang et al.

The aim of this paper is to demonstrate the efficacy of using Contrastive Random Walk as a curiosity method to achieve faster convergence to the optimal policy.Contrastive Random Walk defines the transition matrix of a random walk with the help of neural networks. It learns a meaningful state representation with a closed loop. The loss of Contrastive Random Walk serves as an intrinsic reward and is added to the environment reward. Our method works well in non-tabular sparse reward scenarios, in the sense that our method receives the highest reward within the same iterations compared to other methods. Meanwhile, Contrastive Random Walk is more robust. The performance doesn't change much with different random initialization of environments. We also find that adaptive restart and appropriate temperature are crucial to the performance of Contrastive Random Walk.

Henrik Brådland, Morten Goodwin, Per-Arne Andersen et al.

Document chunking fundamentally impacts Retrieval-Augmented Generation (RAG) by determining how source materials are segmented before indexing. Despite evidence that Large Language Models (LLMs) are sensitive to the layout and structure of retrieved data, there is currently no framework to analyze the impact of different chunking methods. In this paper, we introduce a novel methodology that defines essential characteristics of the chunking process at three levels: intrinsic passage properties, extrinsic passage properties, and passages-document coherence. We propose HOPE (Holistic Passage Evaluation), a domain-agnostic, automatic evaluation metric that quantifies and aggregates these characteristics. Our empirical evaluations across seven domains demonstrate that the HOPE metric correlates significantly (p > 0.13) with various RAG performance indicators, revealing contrasts between the importance of extrinsic and intrinsic properties of passages. Semantic independence between passages proves essential for system performance with a performance gain of up to 56.2% in factual correctness and 21.1% in answer correctness. On the contrary, traditional assumptions about maintaining concept unity within passages show minimal impact. These findings provide actionable insights for optimizing chunking strategies, thus improving RAG system design to produce more factually correct responses.

Adam Štorek, Mukur Gupta, Noopur Bhatt et al.

AI coding assistants are widely used for tasks like code generation. These tools now require large and complex contexts, automatically sourced from various origins$\unicode{x2014}$across files, projects, and contributors$\unicode{x2014}$forming part of the prompt fed to underlying LLMs. This automatic context-gathering introduces new vulnerabilities, allowing attackers to subtly poison input to compromise the assistant's outputs, potentially generating vulnerable code or introducing critical errors. We propose a novel attack, Cross-Origin Context Poisoning (XOXO), that is challenging to detect as it relies on adversarial code modifications that are semantically equivalent. Traditional program analysis techniques struggle to identify these perturbations since the semantics of the code remains correct, making it appear legitimate. This allows attackers to manipulate coding assistants into producing incorrect outputs, while shifting the blame to the victim developer. We introduce a novel, task-agnostic, black-box attack algorithm GCGS that systematically searches the transformation space using a Cayley Graph, achieving a 75.72% attack success rate on average across five tasks and eleven models, including GPT 4.1 and Claude 3.5 Sonnet v2 used by popular AI coding assistants. Furthermore, defenses like adversarial fine-tuning are ineffective against our attack, underscoring the need for new security measures in LLM-powered coding tools.

Kirandeep Kaur, Vinayak Gupta, Aditya Gupta et al.

Most language-based assistants follow a reactive ask-and-respond paradigm, requiring users to explicitly state their needs. As a result, relevant but unexpressed needs often go unmet. Existing proactive agents attempt to address this gap either by eliciting further clarification, preserving this burden, or by extrapolating future needs from context, often leading to unnecessary or mistimed interventions. We introduce ProPer, Proactivity-driven Personalized agents, a novel two-agent architecture consisting of a Dimension Generating Agent (DGA) and a Response Generating Agent (RGA). DGA, a fine-tuned LLM agent, leverages explicit user data to generate multiple implicit dimensions (latent aspects relevant to the user's task but not considered by the user) or knowledge gaps. These dimensions are selectively filtered using a reranker based on quality, diversity, and task relevance. RGA then balances explicit and implicit dimensions to tailor personalized responses with timely and proactive interventions. We evaluate ProPer across multiple domains using a structured, gap-aware rubric that measures coverage, initiative appropriateness, and intent alignment. Our results show that ProPer improves quality scores and win rates across all domains, achieving up to 84% gains in single-turn evaluation and consistent dominance in multi-turn interactions.

Aditya Gupta, Kirandeep Kaur, Vinayak Gupta

In natural language processing, current methods for understanding Transformers are successful at identifying intermediate predictions during a model's inference. However, these approaches function as limited diagnostic checkpoints, lacking a mathematical framework for mechanistically modeling how each layer facilitates transitions between these evolving states. This interpretability gap and past successes of interdisciplinary outlooks inspire us to turn to physics in search of a descriptive mathematical framework for Transformers. We observe that language models are intrinsically probabilistic, an attribute that is echoed in the core postulates of quantum mechanics. This parallel inspires us to translate insights from this discipline to that of natural language processing. Towards this objective, we propose QLENS a novel attempt to develop a physics-based perspective on the Transformer generation process. Under QLENS, a Transformer is studied by converting its latent activations into a state vector in a Hilbert space derived from the model's output units. This state subsequently evolves through hidden layers - reformulated as unitary operators and analogously defined Hamiltonians - during inference. The model's final probability distribution is obtained by applying the Born rule to the end state using a specific measurement operator. To demonstrate QLENS's potential, we conduct a proof-of-concept by probing a toy Transformer to investigate the influence of individual layers in a model's prediction trajectory. We present our work as a foundation for cross-domain insights to be leveraged towards a broader understanding of Transformers.

Mohammad Asif, Diya Srivastava, Aditya Gupta et al.

Inter-subject or subject-independent emotion recognition has been a challenging task in affective computing. This work is about an easy-to-implement emotion recognition model that classifies emotions from EEG signals subject independently. It is based on the famous EEGNet architecture, which is used in EEG-related BCIs. We used the Dataset on Emotion using Naturalistic Stimuli (DENS) dataset. The dataset contains the Emotional Events -- the precise information of the emotion timings that participants felt. The model is a combination of regular, depthwise and separable convolution layers of CNN to classify the emotions. The model has the capacity to learn the spatial features of the EEG channels and the temporal features of the EEG signals variability with time. The model is evaluated for the valence space ratings. The model achieved an accuracy of 73.04%.

Aditya Gupta, Greg Durrett

Tracking entities in procedural language requires understanding the transformations arising from actions on entities as well as those entities' interactions. While self-attention-based pre-trained language encoders like GPT and BERT have been successfully applied across a range of natural language understanding tasks, their ability to handle the nuances of procedural texts is still untested. In this paper, we explore the use of pre-trained transformer networks for entity tracking tasks in procedural text. First, we test standard lightweight approaches for prediction with pre-trained transformers, and find that these approaches underperform even simple baselines. We show that much stronger results can be attained by restructuring the input to guide the transformer model to focus on a particular entity. Second, we assess the degree to which transformer networks capture the process dynamics, investigating such factors as merged entities and oblique entity references. On two different tasks, ingredient detection in recipes and QA over scientific processes, we achieve state-of-the-art results, but our models still largely attend to shallow context clues and do not form complex representations of intermediate entity or process state.

Aditya Gupta, Greg Durrett

Procedural text, which describes entities and their interactions as they undergo some process, depicts entities in a uniquely nuanced way. First, each entity may have some observable discrete attributes, such as its state or location; modeling these involves imposing global structure and enforcing consistency. Second, an entity may have properties which are not made explicit but can be effectively induced and tracked by neural networks. In this paper, we propose a structured neural architecture that reflects this dual nature of entity evolution. The model tracks each entity recurrently, updating its hidden continuous representation at each step to contain relevant state information. The global discrete state structure is explicitly modeled with a neural CRF over the changing hidden representation of the entity. This CRF can explicitly capture constraints on entity states over time, enforcing that, for example, an entity cannot move to a location after it is destroyed. We evaluate the performance of our proposed model on QA tasks over process paragraphs in the ProPara dataset and find that our model achieves state-of-the-art results.