| Page 819 | Kisaco Research

PANEL: Efficient and Sustainable Model Serving for LLMs - Model and Hardware Architectures for Efficient Generative AI Inference

Moderator

Author:

Arun Nandi

Senior Director and Head of Data & Analytics

Unilever

Arun is a visionary AI and Analytics expert recognized as one of the Top 100 Influential AI & Analytics leaders. He is the Head of Data & Analytics at Unilever today. With over 15 years of experience driving analytics-driven value in organizations, he has built AI practices from the ground up on several occasions. Arun advocates the adoption of AI to overcome enterprise-wide challenges and create growth. Beyond his professional achievements, Arun loves to travel, having explored over 40 countries and is passionate about adventure motorbiking.

Panellists

Author:

Nikesh Pahuja

App Engineer

Eximius Systems

Nikesh Kumar Pahuja currently serves as an App Engineer in a contract position at Chevron Corporation, with a primary focus on product development and Data Analytics/ Data Science / AI. With an MBA in Finance, Strategy, and Analytics from the University of California, Davis, and an MS in Analytics from Harrisburg University of Science and Technology, Mr. Pahuja brings a wealth of knowledge and expertise to his work. He is a passionate advocate for data, AI, and the creation of exceptional technological products, driven by a strong commitment to innovation and advancement.

Author:

Jay Dawani

CEO

Lemurian Labs

Jay Dawani is co-founder & CEO of Lemurian Labs, a startup at the forefront of general purpose accelerated computing for making AI development affordable and generally available for all companies and people to equally benefit. Author of the influential book "Mathematics for Deep Learning", he has held leadership positions at companies such as BlocPlay and Geometric Energy Corporation, spearheading projects involving quantum computing, metaverse, blockchain, AI, space robotics, and more. Jay has also served as an advisor to NASA Frontier Development Lab, SiaClassic, and many leading AI firms.

Author:

Praveen Kolli

Staff Machine Learning Engineer

DoorDash

Dr. Praveen Kolli is currently a Staff Machine Learning Engineer in the Ads Quality Team at DoorDash, where he focuses on developing advanced deep learning models to improve ad recommendations and enhance user experience. Prior to his role at DoorDash, he served as a Technical Lead in the Ads Ranking Team at Pinterest. At Pinterest, Dr. Kolli played a pivotal role in building cutting-edge models for personalized ad recommendations, contributing to the platform's success in delivering relevant ads to its users. He holds a PhD in Mathematics from Carnegie Mellon University, Masters in Mathematics of Finance from Columbia University and Bachelors in Electrical Engineering from Indian Institute of Technology, Kharagpur.

Read more about PANEL: Efficient and Sustainable Model Serving for LLMs - Model and Hardware Architectures for Efficient Generative AI Inference

PRESENTATION: Getting Performance and Efficiency from Transformers Without Increasing Parameter Counts - Cutting Edge Techniques in Model Efficiency

The evolution of transformers marks a remarkable advancement in comprehending intricate language patterns. These models are progressively growing in size and precision, enabling them to grasp complex linguistic structures more effectively. However, the real challenge lies in efficiently integrating these advanced models into production systems without incurring exorbitant costs. Striking a balance between superior performance and streamlined resource usage during both training and runtime is crucial for making these large language models (LLMs) viable for deployment in practical production environments.

Author:

Rahul Sharnagat

Principal Machine Learning Engineer

Walmart Global Tech

Rahul Sharnagat is a Principal Data Scientist in Walmart's Search Technologies. He began his career at IBM research lab then worked in Microsoft Bing Ads before joining Walmart in 2018, with his expertise centering around natural language understanding (NER) and question answering. He currently leads the efforts to develop and deploy query understanding models and its application across the search stack. Rahul holds a master's degree in computer science from Indian Institute of Technology, Bombay and a bachelor's degree in computer science from Visvesvaraya National Institute of Technology, Nagpur.

Read more about PRESENTATION: Getting Performance and Efficiency from Transformers Without Increasing Parameter Counts - Cutting Edge Techniques in Model Efficiency

PANEL: Investor Discussion on Enabling Technologies for Enterprise Foundation Model Efficiency - HW Optimization, Model Optimization, MLOps etc.

Moderator

Author:

Michael Stewart

Partner

M12

Panellists

Author:

Rashmi Gopinath

General Partner

B Capital

Rashmi Gopinath is a General Partner at B Capital Group where she leads the fund’s enterprise software practice in cloud infrastructure, cybersecurity, devops, and AI/ML sectors. She brings over two decades of experience investing and operating in cutting-edge enterprise technologies. She led B Capital’s investments in over 24 companies such as DataRobot, FalconX, Clari, Phenom People, Synack, Innovaccer, Labelbox, Fabric, 6Sense, Highspot, Pendo, Starburst, OwnBackup, Figment, Perimeter81, Zesty, among others.

Rashmi was previously a Managing Director at M12, Microsoft’s venture fund, where she led investments globally in enterprise software and sat on several boards including Synack, Innovaccer, Contrast Security, Frame, UnravelData, Incorta, among others.

Prior to M12, Rashmi was an Investment Director with Intel Capital where she was involved in the firm’s investments in startups including MongoDB (Nasdaq: MDB), ForeScout (Nasdaq: FSCT), Maginatics (acq. by EMC), BlueData (acq. by HPE), among others. Rashmi held operating roles at high-growth startups such as BlueData (acq. by HPE) and Couchbase (Nasdaq: BASE) where she led global business development, product and marketing roles. She began her career in engineering and product roles at Oracle and GE Healthcare. She earned an M.B.A. from Northwestern University, and a B.S. in Electrical Engineering from University of Mumbai in India.

Author:

Gayathri Radhakrishnan

Partner

Hitachi Ventures

Gayathri is currently Partner at Hitachi Ventures. Prior to that, she was with Micron Ventures, actively investing in startups that apply AI to solve critical problems in the areas of Manufacturing, Healthcare and Automotive. She brings over 20 years of multi-disciplinary experience across product management, product marketing, corporate strategy, M&A and venture investments in large Fortune 500 companies such as Dell and Corning and in startups. She has also worked as an early stage investor at Earlybird Venture Capital, a premier European venture capital fund based in Germany. She has a Masters in EE from The Ohio State University and MBA from INSEAD in France. She is also a Kauffman Fellow - Class 16.

Author:

Yvonne Lutsch

Investment Principal

Bosch Ventures

Yvonne is an accomplished Investment Principal at Bosch Ventures affiliate office located in Sunnyvale, and sources, evaluates, and executes venture capital deals in North America. Her specialty are investments in deep tech fields such as AI, edge and next gen. computing incl. quantum, robotics, industrial IoT, mobility, climate tech, semiconductors, or sensors. She is an investor and non-executive board member of Bosch Ventures’ portfolio companies Syntiant, Zapata AI, UltraSense Systems, Aclima, and Recogni.
Prior to this position Yvonne was Director of Technology Scouting and Business Development, building up an Innovation Hub in Silicon Valley including startup scouting, business development while advising executives of the Bosch business units on their strategy. She has more than two decades of solid experience in manufacturing operations and engineering in the automotive and consumer electronics space – gained through different executive roles at Bosch in Germany.
Yvonne received a diploma in Experimental Physics from University of Siegen, Germany, and holds a PhD in Applied Physics from University of Tuebingen, Germany.

Author:

Vibhor Rastogi

Global Head AIML Data Investing

Citi Ventures

Read more about PANEL: Investor Discussion on Enabling Technologies for Enterprise Foundation Model Efficiency - HW Optimization, Model Optimization, MLOps etc.

PANEL: Consumption Modalities - Pre-Training vs Finetuning and How to Do Either Easily and Affordably

In this panel, we explore the challenges and opportunities that large, medium, and small companies are facing when they try to pre-train and/or fine-tune Large Language Models (LLMs). More specifically, given the fact that not all companies have the time and resources to pre-train LLMs from scratch, we present options for executing pre-training, or at least fine-tuning, activities in easier and, most importantly, cheaper ways than what is done today. We will therefore discuss how you too can start to efficiently and efficaciously leverage the amazing capabilities provided by LLMs, as well as some important factors that we discovered and that make a significant difference in the creation of high-performance, robust, and reliable LLMs.

Moderator

Author:

Fausto Artico

Head of Innovation and Data Science

GSK

Fausto has two PhDs (Information & Computer Science respectively), earning his second master’s and PhD at the University of California, Irvine. Fausto also holds multiple certifications from MIT, Columbia University, London School of Economics and Political Science, Kellogg School of Management, University of Cambridge and soon also from the University of California, Berkeley. He has worked in multi-disciplinary teams and has over 20 years of experience in academia and industry.

As a Physicist, Mathematician, Engineer, Computer Scientist, and High-Performance Computing (HPC) and Data Science expert, Fausto has worked on key projects at European and American government institutions and with key individuals, like Nobel Prize winner Michael J. Prather. After his time at NVIDIA corporation in Silicon Valley, Fausto worked at the IBM T J Watson Center in New York on Exascale Supercomputing Systems for the US government (e.g., Livermore and Oak Ridge Labs).

Panellists

Author:

Abhijeet Gulati

Head of AI, Senior Director of AI & ML Engineering

Mitchell International

Abhijeet Gulati is the Head of AI & Senior Director of Engineering at Mitchell’s Auto Physical Damage (APD) business unit. Abhijeet is an accomplished technologist with over two decades of experience in the semiconductor, wireless, software & technology industry, focusing on AI, Machine Learning, NLP, Generative AI and SaaS solutions. He is a driven Artificial Intelligence, Advanced Analytics and business intelligence leader.

As the Head of AI at Mitchell International, Abhijeet has spent the past 5 years improving InsureTech inefficiencies, minimize decision biases, developed proprietary, differentiated enterprise scale AI products and an intelligent open platform that democratizes adoption of AI in enterprise business workflows. Abhijeet has extensive experience directing large-scale initiatives involving R&D, business & product strategy, operations, and advanced video, image and data analytics. Abhijeet has authored several patents on the application of AI in InsureTech industry. Abhijeet sits on the board of several AI standards, Ethical & Responsible AI committees.

Author:

Greg Makowski

Head of Data Science Services

Johnson Controls

Author:

Mayank Anand

Machine Learning Engineering Manager

Adobe

Mayank Anand is a Machine Learning Engineering Manager at Adobe, dedicated to enhancing digital marketing with AI and machine learning. His focus lies in harnessing AI for better content creation from texts and images. Presently, he's innovating in the field of generative AI to produce smart, brand-safe content. Mayank holds a Master's in Computer Science from USC, Los Angeles.

Read more about PANEL: Consumption Modalities - Pre-Training vs Finetuning and How to Do Either Easily and Affordably

KEYNOTE: Developer Lens on Achieving Efficient Generative AI with Gaudi(r) Accelerators

Sree Ganesan, who leads Software Products at Intel’s Habana Labs, will present approaches to bringing efficiency to Generative AI—from the software perspective—and will share things you may not know about what you can do with Gaudi Acceleration. We’ll look at opportunities developers have to optimize performance and build value that can yield positive results.

Author:

Sree Ganesan

VP of Product

d-Matrix

Sree Ganesan, VP of Product, d-Matrix: Sree is responsible for product management functions and business development efforts across the company. She manages the product lifecycle, definition and translation of customer needs to the product development function, acting as the voice of the customer. Prior, Sree led the Software Product Management effort at Habana Labs/Intel, delivering state-of-the-art deep learning capabilities of the Habana SynapseAI® software suite to the market. Previously, she was Engineering Director in Intel’s AI Products Group, where she was responsible for AI software strategy and deep learning framework integration for Nervana NNP AI accelerators. Sree earned a bachelor’s degree in electrical engineering from the Indian Institute of Technology Madras and a PhD in computer engineering from the University of Cincinnati, Ohio.

Read more about KEYNOTE: Developer Lens on Achieving Efficient Generative AI with Gaudi(r) Accelerators

KEYNOTE: Homogeneity in Software, Heterogeneity in Compute - Why General Purpose Compute Isn´t Enough for Generative AI

In this keynote, we discuss why commercially available general computing hardware architectures are usually not a good choice for executing innovative AI workloads. We also deep dive into understanding some secrets (no technical expertise needed). Most importantly, we address what we can do to ameliorate this situation. Which type of methodologies and approaches can we create or leverage to efficaciously map new classes of AI algorithms to huge, heterogeneous clusters as well as increase the performance of said algorithms (e.g., accuracy, specificity, etc.)? Which promising and commercially available hardware architectures can we use to quickly and cheaply execute AI workloads? Hint: they are not NVIDIA GPUs. And what could our future software-hardware ecosystems look like, architecturally, to simplify the creation, deployment, execution, and monitoring of AI workloads?

Author:

Fausto Artico

Head of Innovation and Data Science

GSK

Fausto has two PhDs (Information & Computer Science respectively), earning his second master’s and PhD at the University of California, Irvine. Fausto also holds multiple certifications from MIT, Columbia University, London School of Economics and Political Science, Kellogg School of Management, University of Cambridge and soon also from the University of California, Berkeley. He has worked in multi-disciplinary teams and has over 20 years of experience in academia and industry.

As a Physicist, Mathematician, Engineer, Computer Scientist, and High-Performance Computing (HPC) and Data Science expert, Fausto has worked on key projects at European and American government institutions and with key individuals, like Nobel Prize winner Michael J. Prather. After his time at NVIDIA corporation in Silicon Valley, Fausto worked at the IBM T J Watson Center in New York on Exascale Supercomputing Systems for the US government (e.g., Livermore and Oak Ridge Labs).

Read more about KEYNOTE: Homogeneity in Software, Heterogeneity in Compute - Why General Purpose Compute Isn´t Enough for Generative AI

Will Goodhart

Chief Executive Officer

CFA

Rachel Neill

Head of Sustainability

Heptagon Capital

AWC US Sample Attendee List 2023

Karen Wilkinson

Global Head of Climate risk and Net Zero Oversight

Standard Chartered