Saurabh Srivastava

Saurabh Srivastava, Chengyue Huang, Weiguo Fan et al.

Large language models (LLMs) have revolutionized zero-shot task performance, mitigating the need for task-specific annotations while enhancing task generalizability. Despite its advancements, current methods using trigger phrases such as "Let's think step by step" remain limited. This study introduces PRomPTed, an approach that optimizes the zero-shot prompts for individual task instances following an innovative manner of "LLMs in the loop". Our comprehensive evaluation across 13 datasets and 10 task types based on GPT-4 reveals that PRomPTed significantly outperforms both the naive zero-shot approaches and a strong baseline (i.e., "Output Refinement") which refines the task output instead of the input prompt. Our experimental results also confirmed the generalization of this advantage to the relatively weaker GPT-3.5. Even more intriguingly, we found that leveraging GPT-3.5 to rewrite prompts for the stronger GPT-4 not only matches but occasionally exceeds the efficacy of using GPT-4 as the prompt rewriter. Our research thus presents a huge value in not only enhancing zero-shot LLM performance but also potentially enabling supervising LLMs with their weaker counterparts, a capability attracting much interest recently. Finally, our additional experiments confirm the generalization of the advantages to open-source LLMs such as Mistral 7B and Mixtral 8x7B.

Saurabh Srivastava, Annarose M B, Anto P et al.

We propose a framework for robust evaluation of reasoning capabilities of language models, using functional variants of benchmarks. Models that solve a reasoning test should exhibit no difference in performance over the static version of a problem compared to a snapshot of the functional variant. We have rewritten the relevant fragment of the MATH benchmark into its functional variant MATH(), with functionalization of other benchmarks to follow. When evaluating current state-of-the-art models over snapshots of MATH(), we find a reasoning gap -- the percentage difference between the static and functional accuracies. We find reasoning gaps from 58.35% to 80.31% among the state-of-the-art closed and open weights models that perform well on static benchmarks, with the caveat that the gaps are likely to be smaller with more sophisticated prompting strategies. Here we show that models which anecdotally have good reasoning performance over real-world tasks, have quantifiable lower gaps, motivating the open problem of building "gap 0" models. Code for evaluation and new evaluation datasets, three MATH() snapshots, are publicly available at https://github.com/consequentai/fneval/.

Essential AI, Andrew Hojel, Michael Pust et al.

Data plays the most prominent role in how language models acquire skills and knowledge. The lack of massive, well-organized pre-training datasets results in costly and inaccessible data pipelines. We present Essential-Web v1.0, a 24-trillion-token dataset in which every document is annotated with a twelve-category taxonomy covering topic, format, content complexity, and quality. Taxonomy labels are produced by EAI-Distill-0.5b, a fine-tuned 0.5b-parameter model that achieves an annotator agreement within 3% of Qwen2.5-32B-Instruct. With nothing more than SQL-style filters, we obtain competitive web-curated datasets in math (-8.0% relative to SOTA), web code (+14.3%), STEM (+24.5%) and medical (+8.6%). Essential-Web v1.0 is available on HuggingFace: https://huggingface.co/datasets/EssentialAI/essential-web-v1.0

Madhukar Reddy Vongala, Saurabh Srivastava, Jana Košecká

Vision-language pretraining on large datasets of images-text pairs is one of the main building blocks of current Vision-Language Models. While with additional training, these models excel in various downstream tasks, including visual question answering, image captioning, and visual commonsense reasoning. However, a notable weakness of pretrained models like CLIP, is their inability to perform entity grounding and compositional image and text matching~\cite{Jiang2024ComCLIP, yang2023amc, Rajabi2023GroundedVSR, learninglocalizeCVPR24}. In this work we propose a novel learning-free zero-shot augmentation of CLIP embeddings that has favorable compositional properties. We compute separate embeddings of sub-images of object entities and relations that are localized by the state of the art open vocabulary detectors and dynamically adjust the baseline global image embedding. % The final embedding is obtained by computing a weighted combination of the sub-image embeddings. The resulting embedding is then utilized for similarity computation with text embedding, resulting in a average 1.5\% improvement in image-text matching accuracy on the Visual Genome and SVO Probes datasets~\cite{krishna2017visualgenome, svo}. Notably, the enhanced embeddings demonstrate superior retrieval performance, thus achieving significant gains on the Flickr30K and MS-COCO retrieval benchmarks~\cite{flickr30ke, mscoco}, improving the state-of-the-art Recall@1 by 12\% and 0.4\%, respectively. Our code is available at https://github.com/madhukarreddyvongala/GroundingCLIP.

Saurabh Srivastava, Vinay P. Namboodiri, T. V. Prabhakar

AI intensive systems that operate upon user data face the challenge of balancing data utility with privacy concerns. We propose the idea and present the prototype of an open-source tool called Privacy Utility Trade-off (PUT) Workbench which seeks to aid software practitioners to take such crucial decisions. We pick a simple privacy model that doesn't require any background knowledge in Data Science and show how even that can achieve significant results over standard and real-life datasets. The tool and the source code is made freely available for extensions and usage.

Essential AI, Darsh J Shah, Peter Rushton et al.

A language model's ability to reflect on its own reasoning provides a key advantage for solving complex problems. While most recent research has focused on how this ability develops during reinforcement learning, we show that it actually begins to emerge much earlier - during the model's pre-training. To study this, we introduce deliberate errors into chains-of-thought and test whether the model can still arrive at the correct answer by recognizing and correcting these mistakes. By tracking performance across different stages of pre-training, we observe that this self-correcting ability appears early and improves steadily over time. For instance, an OLMo2-7B model pre-trained on 4 trillion tokens displays self-correction on our six self-reflection tasks.

Wenmo Qiu, Saurabh Srivastava

Recent work has explored batch prompting as a strategy to amortize inference cost in large language models (LLMs). In this paper, we show that batching offers an additional, underappreciated benefit: it regularizes model behavior during multi-step reasoning for Large Reasoning Models (LRMs). We conduct a comprehensive study across 13 diverse benchmarks and observe that batching improves accuracy while substantially reducing reasoning token usage, often by 3x-5x. Through detailed behavioral analysis, we find that batching suppresses overthinking, reduces hedging language (e.g., repetitive self-corrections), and encourages more decisive answers. Surprisingly, we also observe emergent collective effects in batched inference: models often generalize patterns from earlier examples to solve harder ones in the same batch. These findings position batching not just as a throughput optimization, but as a powerful inference-time regularizer for more efficient and reliable LLM reasoning.

Essential AI, Ishaan Shah, Anthony M. Polloreno et al.

We demonstrate that Muon, the simplest instantiation of a second-order optimizer, explicitly expands the Pareto frontier over AdamW on the compute-time tradeoff. We find that Muon is more effective than AdamW in retaining data efficiency at large batch sizes, far beyond the so-called critical batch size, while remaining computationally efficient, thus enabling more economical training. We study the combination of Muon and the maximal update parameterization (muP) for efficient hyperparameter transfer and present a simple telescoping algorithm that accounts for all sources of error in muP while introducing only a modest overhead in resources. We validate our findings through extensive experiments with model sizes up to four billion parameters and ablations on the data distribution and architecture.

Saurabh Srivastava, Sweta Pati, Ziyu Yao

In this work, we study the effect of annotation guidelines -- textual descriptions of event types and arguments, when instruction-tuning large language models for event extraction. We conducted a series of experiments with both human-provided and machine-generated guidelines in both full- and low-data settings. Our results demonstrate the promise of annotation guidelines when there is a decent amount of training data and highlight its effectiveness in improving cross-schema generalization and low-frequency event-type performance.

Saurabh Srivastava, Ziyu Yao

Large Reasoning Models (LRMs) such as DeepSeek-R1 and OpenAI o1 have demonstrated remarkable capabilities in various reasoning tasks. Their strong capability to generate and reason over intermediate thoughts has also led to arguments that they may no longer require extensive prompt engineering or optimization to interpret human instructions and produce accurate outputs. In this work, we aim to systematically study this open question, using the structured task of event extraction for a case study. We experimented with two LRMs (DeepSeek-R1 and o1) and two general-purpose Large Language Models (LLMs) (GPT-4o and GPT-4.5), when they were used as task models or prompt optimizers. Our results show that on tasks as complicated as event extraction, LRMs as task models still benefit from prompt optimization, and that using LRMs as prompt optimizers yields more effective prompts. Our finding also generalizes to tasks beyond event extraction. Finally, we provide an error analysis of common errors made by LRMs and highlight the stability and consistency of LRMs in refining task instructions and event guidelines.

Abhishek Dey, Saurabh Srivastava, Gaurav Singh et al.

This paper presents PICO-TINYML-BENCHMARK, a modular and platform-agnostic framework for benchmarking the real-time performance of TinyML models on resource-constrained embedded systems. Evaluating key metrics such as inference latency, CPU utilization, memory efficiency, and prediction stability, the framework provides insights into computational trade-offs and platform-specific optimizations. We benchmark three representative TinyML models -- Gesture Classification, Keyword Spotting, and MobileNet V2 -- on two widely adopted platforms, BeagleBone AI64 and Raspberry Pi 4, using real-world datasets. Results reveal critical trade-offs: the BeagleBone AI64 demonstrates consistent inference latency for AI-specific tasks, while the Raspberry Pi 4 excels in resource efficiency and cost-effectiveness. These findings offer actionable guidance for optimizing TinyML deployments, bridging the gap between theoretical advancements and practical applications in embedded systems.

Saurabh Srivastava, Gaurav Singh, Shou Matsumoto et al.

In this work, we present the first dataset, MailEx, for performing event extraction from conversational email threads. To this end, we first proposed a new taxonomy covering 10 event types and 76 arguments in the email domain. Our final dataset includes 1.5K email threads and ~4K emails, which are annotated with totally ~8K event instances. To understand the task challenges, we conducted a series of experiments comparing three types of approaches, i.e., fine-tuned sequence labeling, fine-tuned generative extraction, and few-shot in-context learning. Our results showed that the task of email event extraction is far from being addressed, due to challenges lying in, e.g., extracting non-continuous, shared trigger spans, extracting non-named entity arguments, and modeling the email conversational history. Our work thus suggests more future investigations in this domain-specific event extraction task.

Hao Yan, Saurabh Srivastava, Yintao Tai et al.

Interactive semantic parsing based on natural language (NL) feedback, where users provide feedback to correct the parser mistakes, has emerged as a more practical scenario than the traditional one-shot semantic parsing. However, prior work has heavily relied on human-annotated feedback data to train the interactive semantic parser, which is prohibitively expensive and not scalable. In this work, we propose a new task of simulating NL feedback for interactive semantic parsing. We accompany the task with a novel feedback evaluator. The evaluator is specifically designed to assess the quality of the simulated feedback, based on which we decide the best feedback simulator from our proposed variants. On a text-to-SQL dataset, we show that our feedback simulator can generate high-quality NL feedback to boost the error correction ability of a specific parser. In low-data settings, our feedback simulator can help achieve comparable error correction performance as trained using the costly, full set of human annotations.

Sumit Soman, Siddharth Srivastava, Saurabh Srivastava et al.

In recent times, there have been significant advancements in utilizing the sensing capabilities of mobile devices for developing applications. The primary objective has been to enhance the way a user interacts with the application by making it effortless and convenient. This paper explores the capabilities of using Brain Computer Interfaces (BCI), an evolving subset of Human Computer Interaction (HCI) paradigms, to control mobile devices. We present a comprehensive survey of the state-of-the-art in this area, discussing the challenges and limitations in using BCI for mobile applications. Further we propose possible modalities that in future can benefit with BCI applications. This paper consolidates research directions being pursued in this domain, and draws conclusions on feasibility and benefits of using BCI systems effectively augmented to the mobile application development domain.