GPT-x, is it really worth it? en

GPT, the state of the art for NLP and NLG. People go crazy in front of the performances. No doubt they impress with their game (well controlled).
Is all that glitters really gold?
An objective analysis of GPT algorithms to discover together how they work under the hood, their strengths, weaknesses, and costs in real life.

Can we outperform GPT-x based models on a budget? en

Trainig a GPT-3 is supposed to cost more than 10 millions of dollars, the cost of their API is not so cheap and they still struggling with bias.

Is it possible to build an architecture able to perform at the same level, in a "budget" ?
Tools like Microsoft LUIS are dead? (Quick answer: No!)
How a good dataset worth 100 times more than an array of GPUs computing massive data.?

Starting from resources like hugging face we will access pre-trained models and libraries that will speed up our job.

Finally we will talk about the final architecture that merges GPT-x with (several) Microsoft LUIS (Basically a QnA maker) in production able to perform as well as very expensive solutions (sometimes far away better) .

Blackmirror - Creating your digital twin en

Starting from a set of sentences describing your personality, thoughts, hobbies, and personal info we will create a digital copy of you to interact with.

The combined architecture of a vanilla version of GPT models and specialized DL models will allow you to perform varied tasks in a cost-effective approach

Moreover, we will create a DL model capable of profiling users in real-time, based on the ongoing conversation with our Digital Twins. The Digital Twins are responsible for acquiring key info that will be extracted by conditioning the flow of the conversation in order to allow you to make important marketing decisions.

There will be a quick demonstration showing you how to utilize this solution to profile live conversations between "humans", or even a youtube video.

Microsoft Azure, as well huggingface, will speed up the whole process

https://ed.ted.com/on/kIvRliGa#digdeeper

AI art and image processing - Unlock your full potential en

In this session we will discover the potential of stable diffusion and the tools available that allows us to create speechless and stunning projects.
Custom training, with our own images is the main focus.

- Dreambooth
- Full training
- Face swapping
- Controlnet
- inpainting

Ethical issues is the final and maybe the most important topic of the session.

Creating your own LLM from opensource models en

From "Simple" Fine-Tuning to your own Mixture of Expert Model Using Open Source Models

Nowadays, training a large language model (LLM) from scratch is a huge effort, even for very large companies. Starting from pre-trained models to create your own custom models is no longer just an option for resource-constrained organizations; it has become a necessary starting point for many.

In this context, various techniques and strategies can help to maximize the potential of pre-trained models:

- Lora: A technique for low-rank adaptation, which allows for efficient fine-tuning of models by focusing on adjusting a small subset of the model's parameters.
- Quantization and QLora: Methods to reduce the computational complexity and memory footprint of models without significantly compromising their performance, enabling more efficient deployment and fine-tuning.
- Managing Multiple Lora Adapters: This involves using multiple Lora adapters to equip models with multiple skills, allowing for a flexible and modular approach to model capabilities.
- Fine Embeddings Management to Improve RAG (Retrieval-Augmented Generation): Enhancing the management of embeddings can significantly improve the performance of RAG systems, which combine the strengths of information retrieval and generative models.
- Mixing Models: Creating Your MoE (Mixture of Experts) Model: This advanced technique involves combining several fine-tuned models to create a Mixture of Experts model, leveraging the strengths of each individual model to enhance overall performance.

These strategies provide a robust toolkit for those who plan to adapt and to enhance pre-trained models to meet specific needs, even without deep expertise in machine learning. By understanding and applying these techniques, organizations can harness the power of modern AI with greater efficiency and effectiveness cutting costs.

Machine Learning - Best Practices and Vulnerabilities en it

Artificial Intelligence and Machine Learning are a must nowdays.
For projects carrying a simple or well-known problem we can find a lot of ready-made solutions, but the game changes when facing with specific custom problems.
The first part of this session is a deep down on techiniques approaches and best practices in configuring ML algorithms but much more, do we really need it always?
The second part will cover vulnerabilities of ML, discovering how easy could be to fool and to hack a neural network by some techniques (Like pixel attack) and their implication in (our) security.

A demostration will focus on a case of e-commerce using cloud ML (Cognitive) services, breaking them then possible solutions and workarounds.

Machine Learning - Best Practices e Vulnerabilità en it

Sempre più progetti oggi necessitano della potenza dell'intelligenza Artificiale ed del Machine Learning.
Per progetti la cui problematica è ben nota si possono trovare soluzioni pronte all'uso, ma la situazione può cambiare radicalmente quando dobbiamo risolvere problematiche specifiche o custom.
Nella prima parte della sessione scopriremo in dettaglio tecniche, approcci e best practices nell'uso e configurazione del ML, ma soprattutto, ne abbiamo sempre realmente bisogno?
Nella seconda parte parleremo di sicurezza e vulnerabilità del Machine Learning scoprendo quanto facile può essere ingannare, hackerare una rete neurale attraverso alcune tecniche (Come il pixel attack) e la loro implicazione nella sicurezza (nostra e dei progetti in azienda).

Verrà fatta una demo su un caso di e-commerce che usa servizi cloud di ML, hackerandoli e quindi possibili soluzioni e workaround.

Deep dive on creating a photorealistic talking avatar en it

Creating a photorealistic avatar speaking any sentence starting from a written input text.

Focusing on autoencoders, we will do a journey from the beginning (Of the speaker experience), mistakes and tips learned along the path.
Will be showcased:

- Intro, the timeline from beginning to nowadays
- Is NOT a deepfake
- Audio processing techniques: STFT (Short Term Fourier Transform), MELs and custom solutions
- Deeplearning models and architecture
- The technique, inspired to inpaiting, used to animate the mouth
- Masks and convolution
- Landmarks extraction
- Morphing animation technique based on autoencoders features
- Microsoft Azure Speech services used to support audio and animation processing
- Putting all together

https://1drv.ms/v/s!AvcsRrhl_mWp-r8WWjc4edmWzZ5tqQ?e=DRHaFK

Creare una avatar foto realistico en it

Dopo un'introduzione a Microsoft Azure Machine Learning Studio, si passerà alla teoria degli algoritmi generativi nel deep learning.

Focalizzandoci sugli autoencoders, verrà proposto un percorso dall'inizio (Dal punto di vista dell'esperienza dello speaker), evidenziando errori principali e suggerimenti imparati nel corso del tempo mostrando diversi modelli di deep learning che hanno portato alla defizione del modello attuale.

https://1drv.ms/v/s!AvcsRrhl_mWp-r8WWjc4edmWzZ5tqQ?e=DRHaFK