Elamur Research⌕
☰
home / authors / Sharan Narang

Sharan Narang

Sharan Narang works on large neural language models and on how their behaviour changes as models and datasets grow. Their papers here return to a few linked questions: how predictably accuracy improves with more data and parameters, how pretraining objectives and architectures compare under controlled conditions, and how the way a model generates text affects its reasoning. The typical method is large-scale empirical study, training many models or a single very large Transformer and measuring the results across varied benchmarks.

The most cited work is Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer. It recasts every language task as mapping input text to output text and uses that shared setup to compare pretraining objectives, architectures, unlabeled corpora and fine-tuning strategies. It also introduces the C4 web corpus. Self-Consistency Improves Chain of Thought Reasoning in Language Models proposes a training-free decoding method. It samples many reasoning paths and keeps the answer most of them agree on. PaLM: Scaling Language Modeling with Pathways describes training a very large dense decoder-only model and tests its few-shot ability across language, code and multilingual tasks. An earlier study, Deep Learning Scaling is Predictable, Empirically, fitted learning curves for speech recognition, image classification, language modeling and translation. It showed that the exponents predicted by classical theory do not match real deep networks.

The newest work turns the scaling question toward model design. Scaling Laws vs Model Architectures examines whether the assumptions a design encodes, its inductive bias, alter how the model scales, both during pretraining and once finetuned. It questions the habit of judging new architectures at one model size only.

Narang's papers here carry bylines from Google between 2019 and 2023, and earlier from Baidu in 2017. Co-authors who recur include Aakanksha Chowdhery, Hyung Won Chung, Adam Roberts and Yanqi Zhou. The work falls mainly under Language models & pretraining, with a strand on reasoning.

Sharan Narang has 5 papers in our library, in 2017–2023, most of them on Language models & pretraining, Reasoning; often with Aakanksha Chowdhery, Adam Roberts, Hyung Won Chung.

5 papers in our library · 2017–2023

OpenAlex · h-index 22 · 43 works · 14 340 citations OpenAlex, 2026-10-14

5 works in our library · 13 067 citations in our library · 3 more works in our library · ⭐ Paper Star rank #125

The profile is the library's match of the name; two people with one name are not always told apart.

Elsewhere: Google Scholar · Semantic Scholar · dblp · ACL Anthology

Affiliations

From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 3 more institutions on single bylines are not listed.

Topics

Language models & pretraining2Reasoning1

Works by year

2020s

20231 paper20222 papers

2010s

20191 paper20171 paper

Papers

The most cited first (OpenAlex and our library).

  1. Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer (2019 · 7 433 citations)
    The authors from Google propose a unified view of transfer learning in NLP. They cast every language task as the conversion of input text into output text: translation, classification, question answering, summarization. This makes it possible to use a single model, a single loss function and the same hyperparameters…
  2. Self-Consistency Improves Chain of Thought Reasoning in Language Models (2022 · 3 035 citations)
    Chain-of-thought prompting helps large language models with multi-step reasoning, but it usually decodes just one greedy path, so a single mistake in that path makes the final answer wrong.
  3. PaLM: Scaling Language Modeling with Pathways (2022 · 2 603 citations)
    The authors investigate how the size of a language model affects its ability to solve tasks from a few examples (few-shot), without task-specific fine-tuning.
  4. Deep Learning Scaling is Predictable, Empirically (2017 · 497 citations)
    Practitioners widely assume that more data and bigger models improve accuracy, but there was no reliable way to forecast how much gain a given increase in training data would bring.
  5. Scaling Laws vs Model Architectures: How does Inductive Bias Influence Scaling? (2023 · 79 citations)
    Most work on scaling laws studies only the standard Transformer. New architectures are also often judged at a single model size, such as "base", on the assumption that their gains carry over to other scales. This paper tests that assumption. The authors ask whether the inductive bias built into an architecture…

Also in our library

3 more works of theirs are in our library without a page here yet; they will appear as we publish them.

Co-authors here

12 of 32 co-authors with a page, the most papers together first.

Cited by

Papers of our library cite their works 13 067 times (of the 140 884 papers whose bibliographies we read); 895 citing papers with a page here →.

This page in Markdown → https://md.research.elamur.ai/free/en/authors/sharan-narang/