This work addresses the need for a balanced approach between performance and efficiency in scalable production environments for visually-rich document understanding (VDU) tasks. Currently, there is a reliance on large document foundation models that offer advanced capabilities but come with a heavy computational burden. In this paper, we propose a multimodal early exit (EE) model design that incorporates various training strategies, exit layer types and placements. Our goal is to achieve a Pareto-optimal balance between predictive performance and efficiency for multimodal document image classification. Through a comprehensive set of experiments, we compare our approach with traditional exit policies and showcase an improved performance-efficiency trade-off. Our multimodal EE design preserves the model's predictive capabilities, enhancing both speed and latency. This is achieved through a reduction of over 20% in latency, while fully retaining the baseline accuracy. This research represents the first exploration of multimodal EE design within the VDU community, highlighting as well the effectiveness of calibration in improving confidence scores for exiting at different layers. Overall, our findings contribute to practical VDU applications by enhancing both performance and efficiency.

Expansion-enhanced sparse lexical representation improves information retrieval (IR) by minimizing vocabulary mismatch problems during lexical matching. In this paper, we explore the potential of jointly learning dense semantic representation and combining it with the lexical one for ranking candidate information. We present a hybrid information retrieval mechanism that maximizes lexical and semantic matching while minimizing their shortcomings. Our architecture consists of dual hybrid encoders that independently encode queries and information elements. Each encoder jointly learns a dense semantic representation and a sparse lexical representation augmented by a learnable term expansion of the corresponding text through contrastive learning. We demonstrate the efficacy of our model in single-stage ranking of a benchmark product question-answering dataset containing the typical heterogeneous information available on online product pages. Our evaluation demonstrates that our hybrid approach outperforms independently trained retrievers by 10.95% (sparse) and 2.7% (dense) in MRR@5 score. Moreover, our model offers better interpretability and performs comparably to state-of-the-art cross encoders while reducing response time by 30% (latency) and cutting computational load by approximately 38% (FLOPs).

The emergence of generative artificial intelligence (GenAI) is transforming information interaction. For decades, search engines such as Google and Bing have been the primary means of locating relevant information for the general population. They have provided search results in the same standard format (the so-called "10 blue links"). The recent ability to chat via natural language with AI-based agents and have GenAI automatically synthesize answers in real-time (grounded in top-ranked results) is changing how people interact with and consume information at massive scale. These two information interaction modalities (traditional search and AI-powered chat) coexist in current search engines, either loosely coupled (e.g., as separate options/tabs) or tightly coupled (e.g., integrated as a chat answer embedded directly within a traditional search result page). We believe that the existence of these two different modalities, and potentially many others, is creating an opportunity to re-imagine the search experience, capitalize on the strengths of many modalities, and develop systems and strategies to support seamless flow between them. We refer to these as panmodal experiences. Unlike monomodal experiences, where only one modality is available and/or used for the task at hand, panmodal experiences make multiple modalities available to users (multimodal), directly support transitions between modalities (crossmodal), and seamlessly combine modalities to tailor task assistance (transmodal). While our focus is search and chat, with learnings from insights from a survey of over 100 individuals who have recently performed common tasks on these two modalities, we also present a more general vision for the future of information interaction using multiple modalities and the emergent capabilities of GenAI.

Automatic summarization is the process of reducing a text document in order to generate a summary that retains the most important points of the original document. In this work, we study two problems - i) summarizing a text document as set of keywords/caption, for image recommedation, ii) generating opinion summary which good mix of relevancy and sentiment with the text document. Intially, we present our work on an recommending images for enhancing a substantial amount of existing plain text news articles. We use probabilistic models and word similarity heuristics to generate captions and extract Key-phrases which are re-ranked using a rank aggregation framework with relevance feedback mechanism. We show that such rank aggregation and relevant feedback which are typically used in Tagging Documents, Text Information Retrieval also helps in improving image retrieval. These queries are fed to the Yahoo Search Engine to obtain relevant images 1. Our proposed method is observed to perform better than all existing baselines. Additonally, We propose a set of submodular functions for opinion summarization. Opinion summarization has built in it the tasks of summarization and sentiment detection. However, it is not easy to detect sentiment and simultaneously extract summary. The two tasks conflict in the sense that the demand of compression may drop sentiment bearing sentences, and the demand of sentiment detection may bring in redundant sentences. However, using submodularity we show how to strike a balance between the two requirements. Our functions generate summaries such that there is good correlation between document sentiment and summary sentiment along with good ROUGE score. We also compare the performances of the proposed submodular functions.

Large language models (LLMs) trained on general domain corpora showed remarkable results on natural language processing (NLP) tasks. However, previous research demonstrated LLMs trained using domain-focused corpora perform better on specialized tasks. Inspired by this pivotal insight, we developed INDUS, a comprehensive suite of LLMs tailored for the Earth science, biology, physics, heliophysics, planetary sciences and astrophysics domains and trained using curated scientific corpora drawn from diverse data sources. The suite of models include: (1) an encoder model trained using domain-specific vocabulary and corpora to address natural language understanding tasks, (2) a contrastive-learning-based general text embedding model trained using a diverse set of datasets drawn from multiple sources to address information retrieval tasks and (3) smaller versions of these models created using knowledge distillation techniques to address applications which have latency or resource constraints. We also created three new scientific benchmark datasets namely, CLIMATE-CHANGE-NER (entity-recognition), NASA-QA (extractive QA) and NASA-IR (IR) to accelerate research in these multi-disciplinary fields. Finally, we show that our models outperform both general-purpose encoders (RoBERTa) and existing domain-specific encoders (SciBERT) on these new tasks as well as existing benchmark tasks in the domains of interest.

During times of crisis, social media platforms play a vital role in facilitating communication and coordinating resources. Amidst chaos and uncertainty, communities often rely on these platforms to share urgent pleas for help, extend support, and organize relief efforts. However, the sheer volume of conversations during such periods, which can escalate to unprecedented levels, necessitates the automated identification and matching of requests and offers to streamline relief operations. This study addresses the challenge of efficiently identifying and matching assistance requests and offers on social media platforms during emergencies. We propose CReMa (Crisis Response Matcher), a systematic approach that integrates textual, temporal, and spatial features for multi-lingual request-offer matching. By leveraging CrisisTransformers, a set of pre-trained models specific to crises, and a cross-lingual embedding space, our methodology enhances the identification and matching tasks while outperforming strong baselines such as RoBERTa, MPNet, and BERTweet, in classification tasks, and Universal Sentence Encoder, Sentence Transformers in crisis embeddings generation tasks. We introduce a novel multi-lingual dataset that simulates scenarios of help-seeking and offering assistance on social media across the 16 most commonly used languages in Australia. We conduct comprehensive cross-lingual experiments across these 16 languages, also while examining trade-offs between multiple vector search strategies and accuracy. Additionally, we analyze a million-scale geotagged global dataset to comprehend patterns in relation to seeking help and offering assistance on social media. Overall, these contributions advance the field of crisis informatics and provide benchmarks for future research in the area.

Modern named entity recognition systems have steadily improved performance in the age of larger and more powerful neural models. However, over the past several years, the state-of-the-art has seemingly hit another plateau on the benchmark CoNLL-03 English dataset. In this paper, we perform a deep dive into the test outputs of the highest-performing NER models, conducting a fine-grained evaluation of their performance by introducing new document-level annotations on the test set. We go beyond F1 scores by categorizing errors in order to interpret the true state of the art for NER and guide future work. We review previous attempts at correcting the various flaws of the test set and introduce CoNLL#, a new corrected version of the test set that addresses its systematic and most prevalent errors, allowing for low-noise, interpretable error analysis.

Clustering-based nearest neighbor search is a simple yet effective method in which data points are partitioned into geometric shards to form an index, and only a few shards are searched during query processing to find an approximate set of top-$k$ vectors. Even though the search efficacy is heavily influenced by the algorithm that identifies the set of shards to probe, it has received little attention in the literature. This work attempts to bridge that gap by studying the problem of routing in clustering-based maximum inner product search (MIPS). We begin by unpacking existing routing protocols and notice the surprising contribution of optimism. We then take a page from the sequential decision making literature and formalize that insight following the principle of ``optimism in the face of uncertainty.'' In particular, we present a new framework that incorporates the moments of the distribution of inner products within each shard to optimistically estimate the maximum inner product. We then present a simple instance of our algorithm that uses only the first two moments to reach the same accuracy as state-of-the-art routers such as \scann by probing up to $50%$ fewer points on a suite of benchmark MIPS datasets. Our algorithm is also space-efficient: we design a sketch of the second moment whose size is independent of the number of points and in practice requires storing only $O(1)$ additional vectors per shard.

This work studies the learnability of unknown quantum circuits in the near term. We prove the natural robustness of quantum statistical queries for learning quantum processes and provide an efficient way to benchmark various classes of noise from statistics, which gives us a powerful framework for developing noise-tolerant algorithms. We adapt a learning algorithm for constant-depth quantum circuits to the quantum statistical query setting with a small overhead in the query complexity. We prove average-case lower bounds for learning random quantum circuits of logarithmic and higher depths within diamond distance with statistical queries. Additionally, we show the hardness of the quantum threshold search problem from quantum statistical queries and discuss its implications for the learnability of shallow quantum circuits. Finally, we prove that pseudorandom unitaries (PRUs) cannot be constructed using circuits of constant depth by constructing an efficient distinguisher and proving a new variation of the quantum no-free lunch theorem.

In this work, we consider the Submodular Maximization under Knapsack (SMK) constraint problem over the ground set of size $n$. The problem recently attracted a lot of attention due to its applications in various domains of combination optimization, artificial intelligence, and machine learning. We improve the approximation factor of the fastest deterministic algorithm from $6+\epsilon$ to $5+\epsilon$ while keeping the best query complexity of $O(n)$, where $\epsilon >0$ is a constant parameter. Our technique is based on optimizing the performance of two components: the threshold greedy subroutine and the building of two disjoint sets as candidate solutions. Besides, by carefully analyzing the cost of candidate solutions, we obtain a tighter approximation factor.