Prajna Upadhyay

Santosh Srinath K, Mudit Somani, Varun Reddy Padala et al.

Modelling a language model for a multi-lingual scenario includes several potential challenges, among which catastrophic forgetting is the major challenge. For example, small language models (SLM) built for low-resource languages by adapting large language models (LLMs) pose the challenge of catastrophic forgetting. This work proposes to employ a continual learning strategy using parts-of-speech (POS)-based code-switching along with a replay adapter strategy to mitigate the identified gap of catastrophic forgetting while training SLM from LLM. Experiments conducted on vision language tasks such as visual question answering and language modelling task exhibits the success of the proposed architecture.

Prajna Upadhyay, Srikanta Bedathur, Tanmoy Chakraborty et al.

Academic search engines allow scientists to explore related work relevant to a given query. Often, the user is also aware of the "aspect" to retrieve a relevant document. In such cases, existing search engines can be used by expanding the query with terms describing that aspect. However, this approach does not guarantee good results since plain keyword matches do not always imply relevance. To address this issue, we define and solve a novel academic search task, called "aspect-based retrieval", which allows the user to specify the aspect along with the query to retrieve a ranked list of relevant documents. The primary idea is to estimate a language model for the aspect as well as the query using a domain-specific knowledge base and use a mixture of the two to determine the relevance of the article. Our evaluation of the results over the Open Research Corpus dataset shows that our method outperforms keyword-based expansion of query with aspect with and without relevance feedback.

Prajna Upadhyay, Tanuma Patra, Ashwini Purkar et al.

In this paper, we describe the construction of TeKnowbase, a knowledge-base of technical concepts in computer science. Our main information sources are technical websites such as Webopedia and Techtarget as well as Wikipedia and online textbooks. We divide the knowledge-base construction problem into two parts -- the acquisition of entities and the extraction of relationships among these entities. Our knowledge-base consists of approximately 100,000 triples. We conducted an evaluation on a sample of triples and report an accuracy of a little over 90\%. We additionally conducted classification experiments on StackOverflow data with features from TeKnowbase and achieved improved classification accuracy.