--- title: "Nitish Shirish Keskar — author profile and papers" description: "Nitish Shirish Keskar: 5 papers on Elamur Research (2017–2021), h-index 23 — affiliations, topics, co-authors and who cites them in our library." canonical: "https://research.elamur.ai/free/en/authors/nitish-shirish-keskar/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/authors/nitish-shirish-keskar/) [home](https://md.research.elamur.ai/free/en/) / [authors](https://md.research.elamur.ai/free/en/authors/) / Nitish Shirish Keskar # Nitish Shirish Keskar Nitish Shirish Keskar works on neural language modelling and natural language processing, with papers here from 2017 to 2021. The work returns to a few questions: how to train and regularise language models, how to steer generated text toward a chosen style, topic or attribute, and how to build and judge systems that handle many tasks at once or condense documents. The methods include recurrent and Transformer models trained on large text collections, benchmarks assembled from public datasets, and close scrutiny of evaluation practice. The papers fall under [language models and pretraining](https://md.research.elamur.ai/free/en/topics/language-models/), summarization, question answering, and bias, fairness and safety. The most cited paper here is [CTRL](https://md.research.elamur.ai/free/en/papers/keskar-2019-ctrl-conditional-transformer-language-model/) (2019). It describes a Transformer with 1.63 billion parameters that always conditions on a control code. The codes come from metadata that web corpora already carry, such as source domain, URL or review rating, so no hand labelling is needed. An earlier paper, [Regularizing and Optimizing LSTM Language Models](https://md.research.elamur.ai/free/en/papers/merity-2017-regularizing-optimizing-lstm-language-models/) (2017), presented AWD-LSTM. It applies DropConnect to the recurrent weights and adds an averaged SGD variant that switches on by itself, all without altering the standard LSTM implementation. [The Natural Language Decathlon](https://md.research.elamur.ai/free/en/papers/mccann-2018-natural-language-decathlon-multitask-learning/) (2018) recast ten tasks, from translation to pronoun resolution, as question answering and proposed a single multitask network to handle them all. The most recent paper, [GeDi](https://md.research.elamur.ai/free/en/papers/krause-2021-gedi-generative-discriminator-guided-sequence/) (2021), carries the interest in controllable generation toward safety and efficiency. It concerns guiding large models such as GPT-2 and GPT-3 away from toxic, biased or negative output. Earlier guided-decoding approaches needed many extra passes for every token, and GeDi uses a generative discriminator to steer decoding instead. A related strand appears in [Neural Text Summarization: A Critical Evaluation](https://md.research.elamur.ai/free/en/papers/kryscinski-2019-neural-text-summarization-critical-evaluation/) (2019). That paper asks why strong news summarisers barely outperform copying the first three sentences, and it examines the datasets, metrics and model outputs behind that result. Bylines on Keskar's papers name Salesforce in San Francisco from 2018 to 2022 and, earlier, Northwestern University from 2015 to 2018. Frequent co-authors include [Richard Socher](https://md.research.elamur.ai/free/en/authors/richard-socher/), [Bryan McCann](https://md.research.elamur.ai/free/en/authors/bryan-mccann/) and [Caiming Xiong](https://md.research.elamur.ai/free/en/authors/caiming-xiong/). There are further collaborations with Stephen Merity and Wojciech Kryscinski. 5 papers on this site · 2017–2021 Nitish Shirish Keskar has 5 papers on this site, in 2017–2021, most of them on [Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/), [Question answering](https://md.research.elamur.ai/free/en/topics/question-answering/), [Summarization](https://md.research.elamur.ai/free/en/topics/summarization/); often with [Richard Socher](https://md.research.elamur.ai/free/en/authors/richard-socher/), [Bryan McCann](https://md.research.elamur.ai/free/en/authors/bryan-mccann/), [Caiming Xiong](https://md.research.elamur.ai/free/en/authors/caiming-xiong/). [OpenAlex](https://openalex.org/A5028709336) · [ORCID](https://orcid.org/0000-0002-2223-8496) · [**h-index 23**](https://openalex.org/authors/A5028709336) · [55 works](https://openalex.org/works?filter=author.id:A5028709336) · [4 998 citations](https://openalex.org/works?filter=author.id:A5028709336&sort=cited_by_count:desc) OpenAlex, 2026-10-14 [5 works on this site](https://md.research.elamur.ai/free/en/authors/nitish-shirish-keskar/#papers) · [4 847 citations in our library](https://md.research.elamur.ai/free/en/authors/nitish-shirish-keskar/cited-by/) · 2 more works in our library The profile is the library's match of the name; two people with one name are not always told apart. Elsewhere: [Google Scholar](https://scholar.google.com/citations?view_op=search_authors&mauthors=Nitish+Shirish+Keskar) · [Semantic Scholar](https://www.semanticscholar.org/search?q=Nitish+Shirish+Keskar) · [dblp](https://dblp.org/search?q=Nitish+Shirish+Keskar) · [ACL Anthology](https://aclanthology.org/search/?q=Nitish+Shirish+Keskar) ## Affiliations - [Salesforce (United States)](https://md.research.elamur.ai/free/en/companies/salesforce/) · San Francisco, United States —2018–2022 on 20 papers - [Northwestern University](https://md.research.elamur.ai/free/en/universities/northwestern-university/) · Evanston, United States —2015–2018 on 7 papers · education (Wikidata) - [OpenAI (United States)](https://md.research.elamur.ai/free/en/companies/openai/) · San Francisco, United States —employment (Wikidata) From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 2 more institutions on single bylines are not listed. ## Topics [Language models & pretraining 2](https://md.research.elamur.ai/free/en/topics/language-models/) · [Question answering 1](https://md.research.elamur.ai/free/en/topics/question-answering/) · [Summarization 1](https://md.research.elamur.ai/free/en/topics/summarization/) · [Bias, fairness & safety 1](https://md.research.elamur.ai/free/en/topics/bias-fairness-safety/) ## Works by year ### 2020s [**2021** 1 paper](https://md.research.elamur.ai/free/en/papers/krause-2021-gedi-generative-discriminator-guided-sequence/) ### 2010s [**2019** 2 papers](https://md.research.elamur.ai/free/en/authors/nitish-shirish-keskar/year/2019/) · [**2018** 1 paper](https://md.research.elamur.ai/free/en/papers/mccann-2018-natural-language-decathlon-multitask-learning/) · [**2017** 1 paper](https://md.research.elamur.ai/free/en/papers/merity-2017-regularizing-optimizing-lstm-language-models/) ## Papers The most cited first (OpenAlex and our library). 1. **[CTRL: A Conditional Transformer Language Model for Controllable Generation](https://md.research.elamur.ai/free/en/papers/keskar-2019-ctrl-conditional-transformer-language-model/)** *(2019 · 1 273 citations)* Large pretrained language models produce fluent text, but users have little influence over its style, topic or purpose beyond writing a prompt. 2. **[Regularizing and Optimizing LSTM Language Models](https://md.research.elamur.ai/free/en/papers/merity-2017-regularizing-optimizing-lstm-language-models/)** *(2017 · 605 citations)* The paper tackles a practical gap in word-level language modeling. Standard regularizers such as dropout and batch normalization transfer poorly to recurrent networks, and many RNN-specific fixes require custom cells that cannot run on fast libraries like NVIDIA cuDNN. The authors build AWD-LSTM, which relies only on… 3. **[Neural Text Summarization: A Critical Evaluation](https://md.research.elamur.ai/free/en/papers/kryscinski-2019-neural-text-summarization-critical-evaluation/)** *(2019 · 555 citations)* Progress in neural text summarization has stalled. Leading systems trained on the CNN/DailyMail news benchmark barely beat the Lead-3 baseline, which simply copies an article's first three sentences. The authors argue that the research setup itself is partly to blame, and they examine its three components: datasets… 4. **[The Natural Language Decathlon: Multitask Learning as Question Answering](https://md.research.elamur.ai/free/en/papers/mccann-2018-natural-language-decathlon-multitask-learning/)** *(2018 · 523 citations)* Most NLP progress comes from models built for one dataset, one metric and one task. The authors argue that general-purpose language systems cannot emerge that way. To test broader models, they propose decaNLP, a benchmark that combines ten public datasets. The tasks are QA, translation, summarization, NLI, sentiment… 5. **[GeDi: Generative Discriminator Guided Sequence Generation](https://md.research.elamur.ai/free/en/papers/krause-2021-gedi-generative-discriminator-guided-sequence/)** *(2021 · 444 citations)* Large language models such as GPT-2 and GPT-3 reproduce their web-crawled training data, including its toxicity, bias and negativity. Steering them toward desired attributes is hard. Discriminator-guided decoding works but is slow: weighted decoding and PPLM need many extra forward or gradient passes per token. ## Also in our library 2 more works of theirs are in our library without a page here yet; they will appear as we publish them. ## Co-authors here 8 co-authors with a page, the most papers together first. - [Richard Socher](https://md.research.elamur.ai/free/en/authors/richard-socher/) ×5 - [Bryan McCann](https://md.research.elamur.ai/free/en/authors/bryan-mccann/) ×4 - [Caiming Xiong](https://md.research.elamur.ai/free/en/authors/caiming-xiong/) ×3 - [Akhilesh Deepak Gotmare](https://md.research.elamur.ai/free/en/authors/akhilesh-deepak-gotmare/) - [Nazneen Fatema Rajani](https://md.research.elamur.ai/free/en/authors/nazneen-fatema-rajani/) - [Shafiq Joty](https://md.research.elamur.ai/free/en/authors/shafiq-joty/) - [Stephen Merity](https://md.research.elamur.ai/free/en/authors/stephen-merity/) - [Wojciech Kryscinski](https://md.research.elamur.ai/free/en/authors/wojciech-kryscinski/) ## Cited by Papers of our library cite their works 4 847 times (of the 140 884 papers whose bibliographies we read); [350 citing papers with a page here →](https://md.research.elamur.ai/free/en/authors/nitish-shirish-keskar/cited-by/).