Artificial intelligence is a pervasive technology that is bringing a sort of revolution in all engineering fields. The Politecnico di Torino, the oldest Italian (poly)technical university, after following for about 160 years of a traditional education and research approach, fully accepts to face this dramatic challenge. This is accomplished by changing deeply the way technologists are educated and innovation is brought to light, so as to keep humans is their fundamental role of decision-makers as long as possible. The transition of engineers from nerds to creative humanists is one of the consequences.

In this presentation, we will introduce a recent proposal in AI that is called Retrieval-Augmented Generation (RAG).
This technique is able to leverage the natural language capabilities of Large Language Model (LLM) such as ChatGPT in order to query a private database. You can literally ask question about your data, without the need to fine-tune an LLM with all your private data. We will show also an example of RAG using Langchain, an open source framework for writing applications for LLM. We will execute a demo with a few lines of Python to chat with private data using a vector database like Chroma or Elasticsearch.

In the era of AutoML (Automated Machine Learning) manually defining a model, training it & deploying it may seem an outdated approach. However, having full control of the model from its definition to its deployment on the cloud allows a fine-grained, production-oriented, and less expensive approach.

Instead of letting the AutoML engine decide everything about the model and the hardware to use, we can do it by ourselves. This approach allows the experienced machine learning practitioner to have full control of the pipeline from the model definition to the hardware used for training and deploying.

In this talk, we'll see how to use TensorFlow Decision Forest (TF-DF) to define a simple regression tree, create a custom Docker image for its training, and train it using Vertex AI, the Google Cloud Machine Learning platform.
The trained model will then be deployed on Google Cloud and it will be ready to accept prediction requests. Eventually, the definition and usage of a client is shown.

Instead of doing all the steps using the web interface, we'll do almost everything using the not-so-known Go client for Vertex AI.

In the current technological landscape, Large Language Models (LLMs) have taken center stage, offering a potent way to infuse intelligence into applications. This presentation delves into the fundamentals and strategies for seamlessly integrating LLMs into your existing applications, enabling them to tap into enhanced capabilities.

During this talk we will explore, in a very practical way, the essential principles and techniques for integrating LLMs into your applications, regardless of programming language or specific libraries.

By taking a technology-agnostic approach, we aim to provide a comprehensive understanding of the core concepts and the boundaries.

In this talk we'll integrate an LLM with the data of a local application understanding all the steps involved.

Artificial Intelligence (AI) offers more efficient ways to implement processes formerly carried out with Data Processing (DP). However, AI has often been used in an ad hoc way. Machines use AI to perform extremely complex functions, but the value of the result depends on the training data sets, which are typically known only to the implementer. In information services, this data issue may have potentially devastating social impacts due to expected bias in the training, but in other applications, such as in autonomous vehicles, the issue is the inability to trace processes leading to a particular decision.
Data Processing standards have played a major role in promoting use of digital technologies for products, services, and applications. However, few examples are known of AI standards with an approach comparable to that of DP standards. MPAI (Moving Picture, Audio, and Data Coding by Artificial Intelligence) has taken on the mission of developing AI-based data coding standards. The group has already developed several Technical Specifications using AI Modules (AIMs) that attempt to break monolithic applications into components with known functions and interfaces and implementable using AI or DP technologies: connected autonomous vehicles, audio enhancement, prediction of company performance, multimodal human-machine conversation, metaverse architecture, neural network watermarking], and portable avatars.
By incorporating these modules (AIMs), applications can be implemented as AI Workflows (AIWs), themselves with known functions and external interfaces, composed of interconnected AIMs.
MPAI Technical Specifications offer the ability to implement AI applications whose operation is more traceable and explainable and the ability to create a competitive market of components – AIMs – with standardised functions and interfaces and potentially providing competitive performance. As an organisation, MPAI offer a rigorous though fast route to standardisation: less than a year for a technology standard.

Today's tech world is rapidly evolving. Merging Artificial Intelligence (AI) with cloud computing has turbocharged data management and decision-making. Couchbase, a database leader, brings you Capella iQ—a groundbreaking AI cloud service transforming how developers use cloud data.
Capella iQ is a game-changer, using AI to boost performance and efficiency in cloud applications. It's like a smart assistant helping developers optimise apps by leveraging data effectively, automating insights, and enhancing overall performance.
Join us as we unravel the basics and the vast potential of Capella iQ. Learn how it catalyses AI-driven cloud services, simplifying data interpretation for developers.
Embark on this journey to comprehend how Capella iQ, with its AI capabilities, reshapes cloud services—ushering in a smarter, more efficient era of data management and utilisation.

Multimodal Large Language Models (MLLMs) are becoming a buzzing term in the machine learning community, thanks to their ability to seamleassy handle a huge variety of input modalities. In essence, these models leverage the knowledge power of Large Language Models (LLMs) to bridge text with images/video, speech etc. This talk aims to unfold the vibrant world of MLLMs, showcasing how they're revolutionising the field of multimodal deep learning.
The spotlight of our discourse will be the journey of MLLMs from multimodal understanding to multimodal generative capabilities. We'll delve into the recently published the Joint Autoregressive Mixture (JAM) framework. This framework marries text and image generation models in a unique model, bringing to the table a unified architecture that excels in high quality interleaved mixed-modal generation. Through a blend of theory, real-world applications, and a sneak peek into the future, this talk is aimed to explore how the world of machine learning is being reshaped with the advent of MLLMs and what this means for the broader pursuit of Artificial General Intelligence.

Explore the world of AI applied to sports, where we showcase how AI-based solutions are revolutionising the viewing experience, fan engagement, coaching & referee support, and operations behind the scenes. Dive with us into the latest advancements in AI technologies such as deep learning, multimedia analysis, generative AI and information retrieval: we will explore the cutting-edge innovations that are shaping the next generation of sports broadcasting and entertainment.

In this presentation, we will explore synthetic data's role in improving the performance of machine learning algorithms. We begin with a primer on synthetic data generation, outlining methodologies for generating high-fidelity datasets. The discussion will then pivot to the creation of synthetic data to bolster model accuracy, robustness, and to address critical concerns such as bias and fairness. Concluding with a discussion on its application, attendees will leave with a practical understanding of the benefits, hurdles, and best practices associated with the deployment of synthetic data in real-world scenarios.

How can we use AI as a superpower without affecting our abilities for logic, social interaction, and collaboration?
We will start with this question, which arose in each of us developers the moment AI was no longer a service, but a tool, frighteningly close to simulating our capabilities.

In this talk we will talk about teamwork and generative AI, and try to put them together responsibly. We will look for an approach and drop it within the scenario of a team made up of humans.

We will list some tools available for developers and try to draw a set of them that can help and speed up work, without changing the way we work, trying to understand how we can use these tools as a virtual pair programmer and not become dependent on them, without losing control.

At least for the time being.

We will explore the state-of-the-art of generative AI for various use cases: after providing an overview of the most renowned services by OpenAI, Microsoft, Midjourney, ElevenLabs and other providers, we will present potential approaches for implementing similar services with open-source tools, frameworks, and models, while leveraging on-premise and/or cloud computational power.

MAX 25 participants

The talk will present an activity called FED-XAI, which is the approach of combining Federated Learning (FL) of eXplainable Artificial Intelligence (AI) models (XAI), in order to improve user experience of the offered communication services by helping end-users trust that in-network AI performs accurate decisions.
Dario will present the FEDXAI demo prototype developed in the framework of the Hexa-X project in collaboration with other partners, and using also §Intel OpenFL framework. Finally, the Intel vision for the future AI-native and cloud-native 6G systems will be provided.

In the rapidly evolving domain of artificial intelligence, open-source Large Language Models (LLMs) are playing a pivotal role. This talk aims to examine the significance of open-source models in democratising AI technology and fostering innovation. We will explore the key players in the open-source LLM landscape, highlighting how these models vary in terms of architecture, capabilities, and applications. Furthermore, practical insights on how to leverage these models for various use cases will be provided. Attendees will gain an understanding of the potential outcomes and transformative impacts that can be achieved through the effective utilisation of open-source LLMs.

The session elucidates the methodology to develop conversational agents by using an internal knowledge base, with a primary focus on the following key facets:
- Designing the solution to align with the foundational principles of AI engineering, concurrently harnessing the conversational proficiencies of the Language Model (LLM).
- Crafting the solution within the framework of a serverless cloud infrastructure.
- Systematically engineering the solution to ensure scalability within a production environment.

In particular, the session is meant to provide the perspective of BIP xTech on creating AI-based assets such that they are sustainable and effective in production with a massive amount of users.

Furthermore, we demonstrate how to integrate the powerful LLMs from OpenAI and Google into an efficient architecture capable of achieving scalability, robustness, and cost-effectiveness while simultaneously matching the expectations of end users with a user-friendly experience.