AI & Generative AI Summit @Amdocs

Join us for an enlightening panel discussion on the role of Generative AI (GenAI) in transforming the enterprise landscape. Moderated by Moshe Freidman (Head of Amdocs Ventures), this session will feature industry leaders and innovators who will discuss their firsthand experiences with implementing GenAI technologies. Learn about the journey from concept to real-world applications, the challenges faced, and the strategies that led to successful deployments. This panel is a must-attend for anyone interested in the practical aspects of deploying GenAI in the real world.

This session is designed for attendees looking to understand the real-world implications of GenAI, seeking strategies to overcome industry-specific challenges, and eager to learn from the experiences of pioneers in this rapidly evolving field. Whether you are involved in startups, corporate strategy, or tech innovation, this discussion will provide valuable insights into effectively leveraging GenAI in real enterprises.

Add to Calendar

In modern enterprises many employees rotinely have to perform tasks related to text understanding and basic manipulation. These include e.g. classification, routing, data extraction into spreadsheets/databases, and formatted summarization into a report. Typically, LLM based approaches to these tasks aim to 'train' or 'fine tune' the LLM to do the same, based on curated labeled data by the organisation itself, and/or external open source datasets from the relevant industry. However, the preparation of such datasets is an expensive, time consuming process. Another issue is that in cases the tuned LLM makes mistakes, it is not clear how many and which new labeled data is required to solve the issue. This makes actual commercial use difficult for many use cases.

Humans, of course, can accomplish the same tasks by being given a written or oral policy. In the NEC research labs, we have created an LLM based process of converting policies into 'prompt ensembles' that are used to effectively 'program' the existing, untuned or lightly tuned LLM to give discrete, constrained-by-prompt answers which are then aggregated and filtered to yield the final result. Constraining the results by breaking up the policy to a 'prompt ensemble' prevents the problem of hallucinations as the LLM answers are discrete or very short. At the same time, the accuracy level is increased by using partially overlapping/repeating prompts. When the resulting system does not implement the policy as desired, or the corporate policy itself is changed, it is easy to locate and modify/add the relevant prompts, and verify the mistake will not be repeated.

We will show sample cases where we have successfully employed this method to several common problems: e.g. fine text classification of emails on the Enron dataset, and compliance verification on contracts and budget related corporate texts. The process is not strongly coupled to the use of a specific LLM, nor to a specific language. As LLMs become stronger, our method directly benefits from that.

Add to Calendar

As the cost of running AI models for training and inference continues to soar, customers are increasingly seeking ways to regain control over their infrastructure expenses. The AI infrastructure landscape is evolving rapidly, from silicon to applications, making traditional methods of managing and automating AI infrastructure untenable.

This is where we can use Gen-AI to run AI:

In this session, we will explore an innovative approach to leveraging Generative AI (Gen-AI) to streamline the management and automation of the AI stack, from initial requirements through to full production. The session will also include a short demo that illustrates how all this works.

Add to Calendar

Recent work suggests that LLM-Based Multi-Agent Systems (LLMAs) can significantly improve performance over single LLM in applications such as commonsense reasoning tasks, defense against jailbreak attacks, urban planning, intelligent personal assistants, gaming, and software development tasks. However, this promising emerging technology still faces many open challenges.  These include decomposing to subtasks, developing role-playing capabilities, the complexity of coordinating multiple agents, scalability issues as the number of LLM agents grows, and the difficulty of maintaining consistent context across the LLM agents. In this talk, we will explore the latest advancements in LLMAs and discuss the key challenges that must be overcome to make this technology widely adopted.

Add to Calendar

Collaboration between edge devices can significantly enhance machine learning (ML) by leveraging vast amounts of data. Federated learning (FL) allows devices to learn from each other without sharing private data, but varying data distributions (non-IID) among devices can reduce accuracy. We  explore how data sharing among users can counter this issue. Data sharing may occur naturally on social graphs or be incentivized by platforms. Our experiments with various ML models and datasets, demonstrate that modest data sharing among neighbors improves learning performance, even in non-IID and IID cases. This shows that data sharing is a simple, effective method for enhancing FL without relying on a central server.

Authors: Mahran Jazi and Irad Ben-Gal

Add to Calendar

I will present the results of the Israeli benchmark and compare it to world data

Add to Calendar

Deep neural networks provide unprecedented performance gains in many real-world problems in signal and image processing. Despite these gains, the future development and practical deployment of deep networks are hindered by their black-box nature, i.e., a lack of interpretability and the need for very large training sets.

On the other hand, signal processing and communications have traditionally relied on classical statistical modeling techniques that utilize mathematical formulations representing the underlying physics, prior information and additional domain knowledge. Simple classical models are useful but sensitive to inaccuracies and may lead to poor performance when real systems display complex or dynamic behaviour. Here we introduce various approaches to model based learning which merge parametric models with optimization tools and classical algorithms leading to efficient, interpretable networks from reasonably sized training sets.  We will consider examples of such model-based deep networks to image deblurring, image separation, super resolution in ultrasound and microscopy, efficient communication systems, and finally we will see how model-based methods can also be used for efficient diagnosis of COVID19 using X-ray and ultrasound.

Add to Calendar

Web technologies like Wasm, WebGL, WebGPU and WebNN today enable running generative models in the browser on the client side, with good performance and sometimes even in real time. We will review the state of the art in applying effects to video, and in image and text generation.

We will demonstrate.

Add to Calendar

As AI personal assistants rapidly advance, we foresee a future where they play an integral role in assisting with everyday tasks, making it imperative for enterprises deploying GenAI applications to manage these interactions with utmost transparency and precision. In an era where the distinction between human users and AI-driven assistants can become increasingly blurred, this session underscores the critical importance of maintaining clear boundaries within these applications.

We will explore strategies for identifying AI interactions, emphasizing the implementation of indicators that enable GenAI systems to accurately detect whether they are engaging with a human or another AI assistant. We will share insights into practical use cases where this distinction is vital, such as customer service scenarios involving billing inquiries, ensuring that these systems not only operate efficiently but also uphold the trust of end users.

Add to Calendar

The process of generating a scoping document given a customer requirement requires deep business and technical understanding of application functionality, processes, and eco-system. Collecting, encoding, searching, retrieving, and iteratively processing scattered, multi-modal information to achieve such comprehensive and accurate scope definition offers compelling challenges. Join us as we share the progress made and lessons learned in Amdocs Global Services while using GenAI technology

Add to Calendar

Search as we know it is broken. This presentation will explore how vector search and GenAI are redefining search technology by enabling more precise, context-aware results. We’ll dive into the mechanics of vector search, how it translates complex user queries into meaningful results, and why it's essential for improving search performance in real-world applications. If you’re looking to enhance your search functionality, this talk will provide the insights and practical examples you need.

Add to Calendar

Join us at the AI Summit to explore how Amdocs is transforming the full testing cycle with our groundbreaking Amdocs Generative AI (amAIz) platform. Our Amdocs Quality Engineering (AQE) framework harnesses the power of generative AI to automate and supervise testing cycles, ensuring faster time to market and superior quality. Imagine effortlessly loading user journey documents and automatically generating comprehensive test cases—our platform makes this a reality.

Add to Calendar

Text-to-image generative diffusion models have recently risen to prominence, achieving unprecedented success and popularity. These text-to-image models are trained on huge amounts of web-scraped image-caption pairs. As a result, these models encode real-world information and correlations, such as the identity of the President of the United States, or the color of the sky. While this knowledge can be useful, and allows easy and efficient generation of beautiful images from simple prompts, it may also be outdated, reflect assumptions and biases (e.g., doctors are always white male), or violate copyrights (as was demonstrated in recent lawsuits for models imitating artistic styles). However, model providers and creators currently have no efficient means to update models without either retraining them---which is costly in computation and time, and might also require data curation---or requiring explicit prompt engineering from the end user.
In this talk, we will discuss three of our recent papers, which aim to offer a fast and practical way to control model behavior post-training. We modify a small, targeted part of the model that is responsible for encoding a certain part in the computation process of the deep network. This is done without training, by editing the model weights using a closed-form solution. The different papers target different parts of the model and various types of information encoded in it: implicit assumptions, factual associations, artistic style, biases, and harmful content. Overall, the methods presented in the talk offer a fast and practical means for safe deployment of text-to-image models.

Add to Calendar

Translating English and other spoken languages into sign language is a complex and nuanced process, shaped by unique grammatical structures, rich morphology, the 3D visual and non linear nature of sign languages and essential cultural considerations. The incorporation of non-manual markers, such as facial expressions and body movements, further complicates the task of accurate translation. This talk will explore these challenges and demonstrate how Generative AI is being applied to address various aspects, enabling more accurate and culturally sensitive sign language translation using avatars.

Add to Calendar

Access to data remains a significant challenge for many organizations due to its high cost and limited availability.  In this session we propose a comprehensive methodology for training and generating data across various formats, including JSON and free text. This methodology aims to facilitate the effective utilization of generated data for applications such as what-if scenarios, performance analysis, and testing.  We address the key challenges encountered throughout the process, from pre-processing to LLM training and data generation. Specifically, we explore techniques for organizing and tokenizing data during pre-processing, strategies for efficient and rapid training, and methods to support diverse data generation use cases.

Add to Calendar

As Large Language Models (LLMs) evolve to process multiple modalities, we face challenges reminiscent of early text-based LLMs: inconsistency, inaccuracy, and hallucinations. This session discusses various techniques for increasing accuracy of vision use cases throught an evaluation based development methodology using AWS AWS Gen AI services, focusing on Anthropic Claude Sonnet 3.5 on Amazon Bedrock.

Add to Calendar