BEGIN:VCALENDAR VERSION:2.0 PRODID:-//wp-events-plugin.com//7.2.3.1//EN TZID:Asia/Kolkata X-WR-TIMEZONE:Asia/Kolkata BEGIN:VEVENT UID:79@cds.iisc.ac.in DTSTART;TZID=Asia/Kolkata:20241104T163000 DTEND;TZID=Asia/Kolkata:20241104T173000 DTSTAMP:20241029T151957Z URL:https://cds.iisc.ac.in/events/cloud-seminar-small-is-beautiful-a-knowl edge-centric-approach-for-small-language-models-dr-manoj-agarwal-nov-4-430 pm-cds-102/ SUMMARY:[CLOUD SEMINAR] Small is Beautiful: A Knowledge Centric Approach fo r Small Language Models\, Dr. Manoj Agarwal\, Nov 4\, 430PM\, CDS 102 DESCRIPTION:\n\nCLOUD COMPUTING SEMINAR SERIES\n\n\n\nTITLE: Small is Beaut iful: A Knowledge Centric Approach for Small Language Models\n\nSPEAKER: M anoj Agarwal\, GiKA.AI\n\nDATE/TIME: Monday\, Nov 4\, 430-530PM\n\nVENUE: CDS 102 Seminar Room\n\n\n\nABSTRACT: LLMs\, although remarkable in genera ting seemingly intelligent answers\, have no real understanding of the dat a they are trained on\, i.e.\, the semantic understanding of the concepts and meanings behind the words (though they “simulate” that understandi ng). Hence\, these models present a few fundamental challenges in their ad option for critical use cases such as:\n\n Hallucination: The hallucinati on is a foundational problem with these models\, even though their world k nowledge is deep and growing.\n Data Leakage: Giant language models have to be called via their APIs and this leaks the private and sensitive data to these centralised large models.\n Limited Context: For many use cases\ , the amount of data to be shared is larger than the context window\, to m ake a meaningful inference. At the same time\, a large context window\, ev en if possible\, may throw the model off track.\n\nBesides\, some of the o ther factors impacting the adoption of these models are lack of up-to-date information\, latency and cost. Some recent approaches\, such as Retrieva l Augmented Generation (RAG)\, are proposed to handle a few of these limit ations. In this talk\, we present a novel approach which is knowledge cent ric instead of document centric\, to address the challenges outlined above . Primarily\, our approach comprises:\n\n1. Understanding the data context : Given a corpus of documents\, it remains a hard question to answer “Wh at is this data about?”. Our first step is to semantically understand th e user data. The data is parsed\, processed\, and cleaned and is used to i mprove the contextual semantic understanding of the small language model u sing a knowledge centric approach.\n\n2. Knowledge Layer: The knowledge la yer is the central component of our system. It captures the data taxonomy and the relationships between the entities and represents the data as know ledge graph. We propose a novel KG-RAG approach\, that is highly effective and efficient to handle the problem of hallucination by grounding the sma ll language model in the factual knowledge while also improving its reason ing capabilities.\n\n3. Domain-Aware Semantic Query Engine: Semantic query engine fetches the factually accurate results by grounding the response i n the data thus facilitating more meaningful\, personalised and contextual search.\n\nBIO: Dr. Manoj Agarwal is a co-founder of a deep tech company\ , GiKA.AI that aims to build next generation data intelligence systems for better semantic search and analytics. Before this\, he was Senior Staff E ngineer in Discovery intelligence team at Uber AI. In Uber\, Manoj introdu ced the semantic search for Uber Eats. Besides\, he worked on automatic en richment of taxonomy using merchant data. Prior to joining Uber\, Manoj wo rked as Principal Applied Scientist at Microsoft – AI and Research and a s a senior researcher in IBM Research. Manoj was the chief architect for b uilding a web scale product knowledge graph for Microsoft – Shopping\, c omprising a few hundred million products. Manoj also worked as adjunct fac ulty in IIT-Gandhinagar. Dr. Manoj Agarwal completed his PhD from IIT-Bomb ay where his thesis was awarded ACM India Doctoral Dissertation Award (Hon orary mention). His research interests are in the areas of web mining\, gr aph mining\, pattern recognition\, data mining\, knowledge graphs\, langua ge models and information retrieval with more than 30 patents and over 25 research paper.\n\nHost: Yogesh Simmhan\n\nAbout: The IBM-IISc Hybrid Clou d Lab (IIHCL) hosted at IISc is curating the Cloud Computing Seminar serie s with guest speakers from Industry and Academia speaking about the latest technologies and research on Cloud and edge computing\, distributed compu ting systems\, and AI/ML/Big Data platforms. More details at: http://iihcl .iisc.ac.in .\n\nALL ARE WELCOME\n\n====================================== ================================ CATEGORIES:Events,Talks END:VEVENT BEGIN:VTIMEZONE TZID:Asia/Kolkata X-LIC-LOCATION:Asia/Kolkata BEGIN:STANDARD DTSTART:20231105T163000 TZOFFSETFROM:+0530 TZOFFSETTO:+0530 TZNAME:IST END:STANDARD END:VTIMEZONE END:VCALENDAR