AI is transforming various sectors, commodities, and fundamental functionalities. Nevertheless, deep neural networks consume excessive resources in terms of memory, computational power, and energy. To ensure the widespread adoption of AI, it must operate efficiently on end-user devices, adhering to strict power and thermal constraints. Techniques such as quantization and compression play a pivotal role in mitigating these challenges.

In this webinar, Sony’s product manager, Amir Servi, will walk you through Sony’s Model Compression Toolkit for quantizing and accelerating deep learning models for efficient edge deployment. You’ll learn how to do the same for your own model! What you’ll learn:

- Our latest research in quantization techniques and its implementation into a practical product

- Importance of hardware-aware compression for inferencing on edge

- How engineers and researchers can implement these techniques through Sony MCT

Recent advancements in computer vision have been significantly propelled by the introduction of transformer architecture and the user-friendly abstractions to pre-train, fine-tune and infer in 🤗 transformers library. This talk provides an overview of the latest transformer-based vision models, explores the utilities available within the 🤗 transformers library, and offers practical insights into the philosophy behind it.

Tired of long lines at retail checkout? Our Intelligent Queue Management system is the answer! Join us for a step-by-step tutorial on how to create such a system using OpenVINO and YOLOv8. We'll walk you through the process of integrating these powerful open-source tools to develop an end-to-end solution that can be deployed in retail checkout environments. You'll learn how to optimize the application to achieve outstanding performance. Whether you're an experienced developer or new to AI, this session will provide practical tips and best practices for building intelligent systems using OpenVINO. By the end of the presentation, you'll have the knowledge and resources to build your own solution.

In an era defined by rapid advancements in artificial intelligence (AI), navigating the ethical landscape of this technology is paramount. In this session, Mónica will unravel the intricate web of ethical dilemmas that accompany AI's transformative power. From addressing bias and fairness to exploring transparency, accountability, and the profound impact of AI on society, Monica will provide insights that shed light on the ethical considerations surrounding AI.

This talk is your opportunity to gain a fundamental understanding of the ethical challenges and responsibilities associated with AI. Mónica will equip you with knowledge that is essential for anyone engaged in AI development, decision-making, or policy formation.

Foundation Models can be demanding in terms of GPU computation and may not be suitable for real-time applications, especially if you want to scale millions of Autonomous Points of Purchase. But we take advantage of the method called knowledge distillation, where we put our foundation models for complex tasks such as annotations and transfer this knowledge into smaller and cost effective models. This allows us to speed up our annotation process up to 90 times faster than human traditional labeling.

Pssst. Want to hear a secret? What if I told you that active learning doesn't have to be hard. What if there were... an easy way? You're in luck. This talk will show you exactly how to implement an active learning pipeline using DagsHub's Data Engine. And 90% of the pipeline can run directly in a Jupyter Notebook or on Google Colab! By the end of the talk, you'll have the information necessary to convert your existing project into one that uses active learning to efficiently and quickly improve your models' metrics!

Using open source tools with YOLOv8 can help you get your next vision AI project up and running, fast. There are repositories of open source images, libraries to help automate data labeling, tools for tracking or counting, and servers for deploying your models. Learn how to use them with YOLOv8 to build your next application.

The ongoing global race for bigger and better artificial intelligence (AI) systems is expected to have a profound societal and environmental impact by altering job markets, disrupting business models, and enabling new governance and societal welfare structures that can affect global consensus for climate action pathways. However, the current AI systems are trained on biased datasets that could destabilize political agencies impacting climate change mitigation and adaptation decisions and compromise social stability, potentially leading to societal tipping events. Thus, the appropriate design of a less biased AI system that reflects both direct and indirect effects on societies and planetary challenges is a question of paramount importance.

Quantization of machine learning (ML) models can lead to significant decrease in the model size as well as reduction in inference latency due to lower bandwidth requirements. When deployed on hardware options that support integer computations efficiently, the performance gains can be even more dramatic. However, quantization can sometimes lead to unacceptable degradation in accuracy. In this talk, we present an overview of the methods to efficiently quantize YOLOv8 models making them an excellent choice for various real-time edge AI applications. We also introduce a class of YOLOv8 models with ReLU6 activation function that show excellent post training quantization results on a variety of model architectures and datasets. Finally, we illustrate how the quantized models can be deployed on multiple hardware options such as CPUs, Edge TPUs, and Orca (DeGirum’s AI HW accelerator) using simple APIs.

Ultralytics is the home for cutting-edge, state-of-the-art computer vision models for tasks like image classification, object detection, image segmentation, and pose estimation. Weights & Biases is a developer-first MLOps platform that when integrated with an Ultralytics workflow, enables us to easily manage our experiments, model checkpoints, and visualize the results of our experiments in an insightful and intuitive manner. In this session, we will explore how we can effectively supercharge our computer vision workflows using Ultralytics and Weights & Biases.

Learn how we created PatentPT, an advanced language model solution that greatly enhances patent search and interaction capabilities. The presentation offers practical insights on fine-tuning and deploying large language models and leveraging enterprise-grade memory agents to autocomplete patents, generate abstracts and claims, and conduct advanced patent search functions using the rich patent corpus. We’ll walk you through how to develop a similar solution using cutting-edge Activeloop’s Deep Lake, the Database for AI, open-source LLM models, Habana Gaudi HPU hardware, and Amazon Sagemaker’s LLM inference APIs.

We will walk you through architectural blueprints and all the steps we took to build the solution – from training our LLM model and finetuning it, creating custom features, and deploying search APIs.

Whether you’re an AI practitioner looking for practical guides on finetuning LLMs, a legal professional interested in leveraging AI for patent search, or simply curious about the future of AI-enhanced solutions, our talk provides a glimpse into the process and potential of using LLMs in a specialized field. Join us as we share our journey of building custom LLM-powered apps powered by Deep Lake, the Database for AI for companies big and small.