AI DevWorld

Tuesday, October 25, 2022

- PDT
PRO Workshop (AI): How Route Optimisation Can Be Scaled and Optimised Using Meta Heuristics for Realistic Scenario
Sushant Burnawal
Sushant Burnawal
Publicis Sapient, Senior Associate Level 1

ECommerce platforms drive the current era, and the COVID pandemic gave rise to the need for home delivery. The end consumers have multiple options to cater for their needs, and in that case, the eCommerce platforms have to provide on-time and quality delivery to stay ahead in the market and, at the same time, boost their profit margins.

Route Optimization is one of the most critical aspects of planning and transportation. It ensures that deliveries always arrive on time and carry out with the lowest possible cost and energy consumption. However, there are a lot of variables that eCommerce platforms need to consider in a real-time scenario.

During this unfortunate COVID pandemic, eCommerce platforms deal with a massive inflow of e-commerce orders from customers scattered throughout a city, country or even across the globe. This gives rise to an enormous number of variables come into play that cannot be solved using conventional methods in a reasonable amount of time. With the recent developments in AI, machine learning and cloud data, the entire game of route optimization has begun to change. AI continuously retrieves data, learns from it, and searches for improved methods to ensure the most optimal routes for the drivers.

In the novel solution, we are trying to solve the multi-objective vehicle routing problem with optimization variables like minimizing the delivery cost, the number of vehicles and delivery time. To show this as a real-life simulation, we will dissect through the open-source library of veroviz combined with innovative scaling solutions to showcase the real-time implementation of route optimization in any part of the world. 

- PDT
PRO Workshop (AI): Sparsity without Sacrifice – How to Accelerate AI Models Without Losing Accuracy
Anshuman Mishra
Anshuman Mishra
Numenta, Principal Researcher
Lawrence Spracklen
Lawrence Spracklen
Numenta, Director of Machine Learning Architecture

Most companies with AI models in production today are grappling with stringent latency requirements and escalating energy costs. One way to reduce these burdens is by pruning such models to create sparse lightweight networks. Pruning involves the iterative removal of weights from a pre-trained dense network to obtain a network with fewer parameters, trading off against model accuracy. Determining which weights should be removed in order to minimize the impact to the network’s accuracy is critical. For real-world networks with millions of parameters, however, analytical determination is often computationally infeasible; heuristic techniques are a compelling alternative.In this presentation, we talk about how to implement commonly-used heuristics such as gradual magnitude pruning (GMP) in production, along with their associated accuracy-speed trade offs, using the BERT family of language models as an example.Next, we cover ways of accelerating such lightweight networks to achieve peak computational efficiencies and reduce energy consumption. We walk through how our acceleration algorithms optimize hardware efficiency, unlocking order-of-magnitude speedups and energy savings.Finally, we present best practices on how these techniques can be combined to achieve multiplicative effects in reducing energy consumption costs and runtime latencies without sacrificing model accuracy.


- PDT
PRO Workshop (AI): Scaling ML Embedding Models to Serve a Billion Queries
Senthilkumar Gopal
Senthilkumar Gopal
eBay, Senior Engineering Manager, Search ML

This talk is aimed at providing a deeper insight into the scale, challenges and solutions formulated for powering embeddings based visual search in eBay. This talk walks the audience through the model architecture, application archite for serving the users, the workflow pipelines produced for building the embeddings to be used by Cassini, eBay's search engine and the unique challenges faced during this journey. This talk provides key insights specific to embedding handling and how to scale systems to provide real time clustering based solutions for users. 

- PDT
PRO Workshop (AI): Artificial General Intelligence with GPT-3 with Open AI
Cameron Vetter
Cameron Vetter
Octavian Technology Group, Principal Architect

Large Language Models (LLM) have come out of the realm of academia and research and become available to average development teams thanks to the efforts of Open AI and their competitors. Now that we have access to them what can we do with them?

This talk will explore some of the practical uses for GPT-3 made available through Open AI. We will start with a brief introduction to LLM's and transformers and how they bring us a step closer to artificial general intelligence. We will focus on real demonstrations. Each capability will start with a canned demonstration and move on to ad hoc input provided by the audience.

• Text Generation
○ Turn complex text into a simple summary
○ Create an outline of an essay
• Conversation
○ Sarcastic chat bot
• Code Generation
○ Explain Python Code
○ Translate text into programmatic commands
• Question Answering
○ Factual Answering

You will leave this talk with an understanding of Large Language Models and their practical use cases. Walk away inspired on how to apply large language models to your business today! 

Wednesday, October 26, 2022

- PDT
PRO TALK (AI): ML Drift Monitoring : What to Observe, How to Analyze & When to Act
Kumaran Ponnambalam
Kumaran Ponnambalam
Cisco, Principal Engineer

Deploying a new ML model in production successfully is a great achievement, but also is the beginning of a persistent challenge to keep them performing at expected levels. Models in product will drift and decay, and the value provided by them to the business will drop. ML drift monitoring is a challenging tasks, from identifying the right data to collect, the right metrics to compute, the right trends to analyze and the right actions to take. This session will explore the process of model drift monitoring, from model instrumentation to determining the next-best-action. Real life challenges will be explored and best practices and recommendations will be discussed. 

- PDT
OPEN TALK (AI): Lessons Learned Building Natural Language Systems in Healthcare
David Talby
David Talby
John Snow Labs, CTO

This session reviews case studies from real-world projects that built AI systems that use Natural Language Processing (NLP) in healthcare. These case studies cover projects that deployed automated patient risk prediction, automated diagnosis, clinical guidelines, and revenue cycle optimization.

We will cover why and how NLP was used, what deep learning models and libraries were used, how transfer learning enables tuning accurate models from small datasets, and what was productized and achived. Key takeaways for attendees will include applicable best practices for NLP projects including how to build domain-specific healthcare models and using NLP as part of larger machine learning and deep learning pipelines. 

- PDT
OPEN TALK (AI): Deep Dive on Creating a Photorealistic Talking Avatar
Sebastiano Galazzo
Sebastiano Galazzo
Synapsia.ai, Artificial intelligence researcher

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 

- PDT
PRO TALK (AI): Data Ecosystem a Stepping Stone for Decarbonization of Operation Industry

Climate change is possibly one of the most complex and challenging issues on earth. On the other hand, manufacturing companies often find themselves in the crosswind of it. Oil and gas, mining, chemical, cement, energy, and utility sectors are responsible for more than 50% of the industrial GHG emissions. The changes they are bringing into their operations are not enough to address the issue. New initiatives for carbon abetment are not showing any visible improvement in reducing GHG levels in the environment.

In this session, we will analyze how data ecosystems such as LiDAR, remote-sensing data, IT, and OT data pertinent to these manufacturing companies can help them to track/measure, trace and mitigate excess emission issues for their operations. We will also explore how advanced AI techniques such as deep learning, and reinforcement learning techniques can be used effectively to find an optimal solution for the above-mentioned problem/s with real-life examples. 

- PDT
PRO TALK (AI): Designing and Applying AI on Large Volume of IOT Telemetry Data in Azure
Rahat Yasir
Rahat Yasir
ISAAC Instrument, AI, Director of Data Science & AI

IOT devices are producing large amount of telemetry data. We need to ingest them, store them, visualize them, analyze them and build ML models on them to make those data useful. In this session, we will talk about the ways to deal with IOT data, using IOT data to train AI models, building AI models and deploying AI models to inference on real time large volume of IOT data using Azure AI tools. 

- PDT
OPEN TALK (AI): How To Build An AI Based Knowledge Graph for Customers in Fintech
Gautam Gupta
Gautam Gupta
Intuit, Technology leader

In this session, we’d go through our journey to build an AI based Customer Knowledge graph. We’d share the insights & knowhow required to create this scalable & polyglot data platform. Join us to learn the design patterns & best practices that we have developed over time to create an intelligent solution based on AI & Graph technologies for an ever increasing list of product lines and customers. 

- PDT
OPEN TALK (AI): Patenting Artificial Intelligence– How AI Companies Can Identify and Protect AI Inventions
Steve Bachmann
Steve Bachmann
Bachmann Law Group PC, President, Silicon Valley Patent Attorney

Artificial intelligence is becoming one of the most widespread and useful technologies in use today. From data collection to model training, language processing to predictive models, deep networks to AI frameworks, there are many categories and implementations of AI, all with protectable features and important business applications. Protecting cutting edge AI technology helps companies achieve business goals and support their AI innovation.
This presentation will identify key strategies to identify which aspects of AI are patentable and which aspects are not. The discussed strategies will be supplemented with practical real-world examples of patenting different areas of the AI process, from data collection to model training and model implementation to output applications, as well as distinct types of AI systems.
Attendees will also learn about AI patent trends and the most common use cases in which different AI companies build valuable patent portfolios around their AI technology. 

- PDT
OPEN TALK (AI): Scalable, Explainable and Unsupervised Anomaly Detection for Telecom
Ivan Caramello de Andrade
Ivan Caramello de Andrade
Encora Brazil Division, Innovation Leader and Tech Lead

In developing and implementing a telecommunications network, one of the most oppressive challenges that these companies deal with are anomalies that occur within the network showing that something strange (usually an attack, a fraud or an error) is happening. Detecting these anomalies is a challenge because they may appear in different places and formats and require the observation of multiple metrics over hundreds of thousands of events to tell regular behaviors from anomalous ones. Ivan Carmello De Andrade, would like to explain how detecting these anomalies with higher accuracy may be possible with the technology and machine learning capabilities of today.

In his technical session, Ivan will explain how he and his team were able to customize and adapt a Robust Random Cut Forest model to identify and explain anomalies in an unsupervised and scalable way. He and his team will explain the process behind creating this solution as well as the challenges they overcame in development, such as extracting behaviors from individual events. He will also explain the benefit of this model to the user which include:

• The user does not need to understand which behaviors are regular or anomalous nor which features are relevant to describe and identify them
• The model provides accountability, because the user can identify and understand which factors lead to an event being identified as an anomaly
• Scalability in general, the model can be implemented on many different scales with a highly distributable structure and configurable levels of detail 

- PDT
OPEN TALK (AI): Pushing Deepfakes to the Limit - Fake Video Calls with AI
Martin Förtsch
Martin Förtsch
TNG Technology Consulting GmbH, Principal Consultant
Thomas Endres
Thomas Endres
TNG Technology Consulting GmbH, Partner
Jonas Mayer
Jonas Mayer
TNG Technology Consulting GmbH, Senior Consultant

Today's real-time Deepfake technology makes it possible to create indistinguishable doppelgängers of a person and let them participate in video calls. Since 2019, the TNG Innovation Hacking Team has intensively researched and continuously developed the AI around real-time Deepfakes. The final result and the individual steps towards photorealism will be presented in this talk.

Since its first appearance in 2017, Deepfakes have evolved enormously from an AI gimmick to a powerful tool. Meanwhile different media outlets such as "Leschs Kosmos", Galileo and other television formats have been using TNG Deepfakes.

In this talk we will show the different evolutionary steps of the Deepfake technology, starting with the first Deepfakes and ending with real-time Deepfakes of the entire head in high resolution. Several live demos will shed light on individual components of the software. In particular, we focus on various new technologies to improve Deepfake generation, such as Tensorflow 2 and MediaPipe, and the differences in comparison to our previous implementations. 

- PDT
OPEN TALK (AI): Democratizing Deep Learning with Vector Similarity Search
Nava Levy
Nava Levy
Redis, AI/ML Developer Advocate

Deep learning is responsible for most of the breakthroughs we have seen in AI/ML in recent years, yet most companies' models in production use classic or traditional ML. In this talk we will explore how deep learning is being democratized today, thanks to the rising use and availability of vector embeddings from giant pre-trained neural networks. We will see how these embeddings can be combined together with vector similarity search to address different use cases covering any modality and applied to any type of object. Finally, we will discuss the many opportunities this presents as well as the tools that are required to successfully deploy these applications into production. 

Thursday, October 27, 2022

- PDT
PRO TALK (AI): Physics-Based Graph Neural Networks Enable Composable, Strongly Typed Neural Networks
Troy Harvey
Troy Harvey
PassiveLogic, Co-founder, CEO, and Product Architect

PassiveLogic’s (www.passivelogic.com) platform for generalized autonomy utilizing Deep Digital Twins is built on systems-level control theory. The platform is generalized because it can be used to control any kind of system. At its core, this type of platform works on the sensor-fusion and control-fusion of digital models. In these Deep Digital Twin models, the digital twin literally is the AI structure. Each digital twin utilizes the fundamentals of physics to model a single component or piece of equipment. When multiple digital twins are linked to each other in a graph neural network, they form a system description. Because their physics are integral to the models themselves, these graph-based system descriptions model not only the real complexities of systems but also their emergent behavior and the system semantics.
Deep physics networks are structured similar to neural networks, but unlike the homogeneous activation functions of neural nets, each neuron comprises unique physical equations representing a function in a thermodynamic system. The Deep Physics approach is built on heterogeneous neural nets that are composable, have physics guarantees, allow users to define their own systems, learn unsupervised, and generate a physics description of a system. Being so principled, it is also necessarily more constrained, meaning the physics-based graph neural networks can be used to predict future system behavior.
The physics-based graph neural network provides a systems-level intelligence as it understands the interconnectivity of components in a system. As such, it can automatically infer behavior and introspect results, even where sensors do not exist. Using this inference ability, an autonomous control platform built on Deep Digital Twins can provide self-commissioning, automate point-mapping, validate installation, and provide continuous system measurement and verification against its original design. Real-time system operational data can be brought into the model for real-time machine learning so that the model can adapt for improved accuracy of predicting the system behavior.
In this talk, Troy Harvey, CEO at PassiveLogic, will describe Deep Digital Twin AI structures and the applications for generalized autonomy. 

- PDT
OPEN TALK (AI): Operationalizing AI with a Shift from Research to Product Orientation
Yotam Oren
Yotam Oren
Mona, CEO & Cofounder

Many AI programs fail to deliver sustained value despite great research, due to insufficient operational tools, processes and practices. These days, more and more data science teams are going through a major shift, from research orientation, to product orientation. Key factors to successfully transition to a product-oriented approach to AI include empowering data scientists to take end to end accountability for model performance, and going beyond the model - gaining a granular understanding of the behavior of the entire AI-driven process. In this talk, Yotam will discuss the importance of empowering data science teams to successfully make the transition from research oriented to product oriented. 

- PDT
OPEN TALK (AI): Level Up Your Data Lake - to ML and Beyond
Vinodhini SD
Vinodhini SD
Treeverse, Developer Advocate

A data lake is primarily two things: an object store and the objects being stored. Even with the most basic setup, data lakes are capable of supporting BI, Machine Learning, and operational analytics use cases. This flexibility speaks to the strength of object stores, particularly their flexibility in integrating with a diverse set of data processing engines.

As data lakes exploded in adoption, a number of improvements were made to the first architectures. The first and most obvious improvement was to file formats, which led to the development of analytics-optimized formats like parquet, and eventually modern table formats.

An even newer improvement has been the emergence of data source control tools that bring new levels of manageability across an entire lake! In this talk, we'll cover how to incorporate these technologies into your data lake, and how they simplify workflows critical to ML experimentation, deployment of datasets, and more! 

- PDT
OPEN TALK (AI): Reducing Latency and Resource Consumption for Offline Feature Generation
Dhaval Patel
Dhaval Patel
Netflix, Machine Learning Infrastructure

Personalization is one of the key pillars of Netflix as it enables each member to experience the vast collection of content tailored to their interests. Our personalization system is powered by various machine learning models. We constantly innovate by adding new features to our personalization models and running A/B tests to improve recommendations for our members. We also continue to see that providing larger training sets to our models helps make better predictions. Our ML fact store has enabled us to provide larger training sets where the training set spans over a long time window. While a great success, the ML fact store architecture has its limitations. For example, features computed while generating recommendations must be recomputed by offline feature generation pipelines. This talk is about those limitations and how we enhanced our architecture to run optimized offline feature generation pipelines. 

- PDT
OPEN TALK (AI): Bringing Life and Motion to AI Explainability
Joao Nogueira
Joao Nogueira
Optum, Senior AI Engineer
Pietro Mascolo
Pietro Mascolo
Optum Ireland, Data Scientist

SHAP is a great tool to help developers and users understand black box models. To push it to the next level, we will show how to leverage on Dash, SHAP, gifs, and auto-encoders to generate interactive dashboards with animations and visual representations to understand how different AI models learn and change their minds while progressively trained with growing amounts of data.

Animations will help developers understand how frequently AI models tweak their population and local importance factors during training and how they compare across competing AI models, adding an extra layer to AI safety. Auto-encoders and LSTM will be used to generate 2-dimensional embedding representations of explainability paths at individual level, allowing developers to interactively detect algorithm decision making similarity across time and visually debug mislabeled AI predictions at each point in time.

We will show this application in the context of Chronic Kidney Disease prediction and broader Healthcare AI. 

Tuesday, November 1, 2022

- PDT
[#VIRTUAL] PRO Workshop (AI): How Route Optimisation Can Be Scaled and Optimised Using Meta Heuristics for Realistic Scenario
Sushant Burnawal
Sushant Burnawal
Publicis Sapient, Senior Associate Level 1

ECommerce platforms drive the current era, and the COVID pandemic gave rise to the need for home delivery. The end consumers have multiple options to cater for their needs, and in that case, the eCommerce platforms have to provide on-time and quality delivery to stay ahead in the market and, at the same time, boost their profit margins.

Route Optimization is one of the most critical aspects of planning and transportation. It ensures that deliveries always arrive on time and carry out with the lowest possible cost and energy consumption. However, there are a lot of variables that eCommerce platforms need to consider in a real-time scenario.

During this unfortunate COVID pandemic, eCommerce platforms deal with a massive inflow of e-commerce orders from customers scattered throughout a city, country or even across the globe. This gives rise to an enormous number of variables come into play that cannot be solved using conventional methods in a reasonable amount of time. With the recent developments in AI, machine learning and cloud data, the entire game of route optimization has begun to change. AI continuously retrieves data, learns from it, and searches for improved methods to ensure the most optimal routes for the drivers.

In the novel solution, we are trying to solve the multi-objective vehicle routing problem with optimization variables like minimizing the delivery cost, the number of vehicles and delivery time. To show this as a real-life simulation, we will dissect through the open-source library of veroviz combined with innovative scaling solutions to showcase the real-time implementation of route optimization in any part of the world. 

- PDT
[#VIRTUAL] PRO Workshop (AI): Sparsity without Sacrifice – How to Accelerate AI Models Without Losing Accuracy
Anshuman Mishra
Anshuman Mishra
Numenta, Principal Researcher
Lawrence Spracklen
Lawrence Spracklen
Numenta, Director of Machine Learning Architecture

Most companies with AI models in production today are grappling with stringent latency requirements and escalating energy costs. One way to reduce these burdens is by pruning such models to create sparse lightweight networks. Pruning involves the iterative removal of weights from a pre-trained dense network to obtain a network with fewer parameters, trading off against model accuracy. Determining which weights should be removed in order to minimize the impact to the network’s accuracy is critical. For real-world networks with millions of parameters, however, analytical determination is often computationally infeasible; heuristic techniques are a compelling alternative.In this presentation, we talk about how to implement commonly-used heuristics such as gradual magnitude pruning (GMP) in production, along with their associated accuracy-speed trade offs, using the BERT family of language models as an example.Next, we cover ways of accelerating such lightweight networks to achieve peak computational efficiencies and reduce energy consumption. We walk through how our acceleration algorithms optimize hardware efficiency, unlocking order-of-magnitude speedups and energy savings.Finally, we present best practices on how these techniques can be combined to achieve multiplicative effects in reducing energy consumption costs and runtime latencies without sacrificing model accuracy.


- PDT
[#VIRTUAL] PRO Workshop (AI): Scaling ML Embedding Models to Serve a Billion Queries
Senthilkumar Gopal
Senthilkumar Gopal
eBay, Senior Engineering Manager, Search ML

This talk is aimed at providing a deeper insight into the scale, challenges and solutions formulated for powering embeddings based visual search in eBay. This talk walks the audience through the model architecture, application archite for serving the users, the workflow pipelines produced for building the embeddings to be used by Cassini, eBay's search engine and the unique challenges faced during this journey. This talk provides key insights specific to embedding handling and how to scale systems to provide real time clustering based solutions for users. 

- PDT
[#VIRTUAL] PRO Workshop (AI): Artificial General Intelligence with GPT-3 with Open AI
Cameron Vetter
Cameron Vetter
Octavian Technology Group, Principal Architect

Large Language Models (LLM) have come out of the realm of academia and research and become available to average development teams thanks to the efforts of Open AI and their competitors. Now that we have access to them what can we do with them?

This talk will explore some of the practical uses for GPT-3 made available through Open AI. We will start with a brief introduction to LLM's and transformers and how they bring us a step closer to artificial general intelligence. We will focus on real demonstrations. Each capability will start with a canned demonstration and move on to ad hoc input provided by the audience.

• Text Generation
○ Turn complex text into a simple summary
○ Create an outline of an essay
• Conversation
○ Sarcastic chat bot
• Code Generation
○ Explain Python Code
○ Translate text into programmatic commands
• Question Answering
○ Factual Answering

You will leave this talk with an understanding of Large Language Models and their practical use cases. Walk away inspired on how to apply large language models to your business today! 

- PDT
[#VIRTUAL] PRO Workshop (AI): Intro to Machine Learning with ML.NET
David Patrick
David Patrick
DSA, MCT, MCSD, MCSE, MVP

Come and get immersed into the world of machine learning with this introduction and demonstration of to ML.NET. We'll show how to create an app that can predict the type of iris flower based on features such as petal length. We'll show how to download and install ML.NET, create a data set, write the required c# code and run the finished app. 

Wednesday, November 2, 2022

- PDT
[#VIRTUAL] PRO TALK (AI): ML Drift Monitoring : What to Observe, How to Analyze & When to Act
Kumaran Ponnambalam
Kumaran Ponnambalam
Cisco, Principal Engineer

Deploying a new ML model in production successfully is a great achievement, but also is the beginning of a persistent challenge to keep them performing at expected levels. Models in product will drift and decay, and the value provided by them to the business will drop. ML drift monitoring is a challenging tasks, from identifying the right data to collect, the right metrics to compute, the right trends to analyze and the right actions to take. This session will explore the process of model drift monitoring, from model instrumentation to determining the next-best-action. Real life challenges will be explored and best practices and recommendations will be discussed. 

- PDT
[#VIRTUAL] OPEN TALK (AI): Lessons Learned Building Natural Language Systems in Healthcare
David Talby
David Talby
John Snow Labs, CTO

This session reviews case studies from real-world projects that built AI systems that use Natural Language Processing (NLP) in healthcare. These case studies cover projects that deployed automated patient risk prediction, automated diagnosis, clinical guidelines, and revenue cycle optimization.

We will cover why and how NLP was used, what deep learning models and libraries were used, how transfer learning enables tuning accurate models from small datasets, and what was productized and achived. Key takeaways for attendees will include applicable best practices for NLP projects including how to build domain-specific healthcare models and using NLP as part of larger machine learning and deep learning pipelines. 

- PDT
[#VIRTUAL] PRO TALK (AI): Data Ecosystem a Stepping Stone for Decarbonization of Operation Industry

Climate change is possibly one of the most complex and challenging issues on earth. On the other hand, manufacturing companies often find themselves in the crosswind of it. Oil and gas, mining, chemical, cement, energy, and utility sectors are responsible for more than 50% of the industrial GHG emissions. The changes they are bringing into their operations are not enough to address the issue. New initiatives for carbon abetment are not showing any visible improvement in reducing GHG levels in the environment.

In this session, we will analyze how data ecosystems such as LiDAR, remote-sensing data, IT, and OT data pertinent to these manufacturing companies can help them to track/measure, trace and mitigate excess emission issues for their operations. We will also explore how advanced AI techniques such as deep learning, and reinforcement learning techniques can be used effectively to find an optimal solution for the above-mentioned problem/s with real-life examples. 

- PDT
[#VIRTUAL] PRO TALK (AI): Designing and Applying AI on Large Volume of IOT Telemetry Data in Azure
Rahat Yasir
Rahat Yasir
ISAAC Instrument, AI, Director of Data Science & AI

IOT devices are producing large amount of telemetry data. We need to ingest them, store them, visualize them, analyze them and build ML models on them to make those data useful. In this session, we will talk about the ways to deal with IOT data, using IOT data to train AI models, building AI models and deploying AI models to inference on real time large volume of IOT data using Azure AI tools. 

- PDT
[#VIRTUAL] PRO Workshop (AI): Deploying Machine Learning Models with Pulsar Functions
David Kjerrumgaard
David Kjerrumgaard
StreamNative, Developer Advocate

In this talk I will present a technique for deploying machine learning models to provide real-time predictions using Apache Pulsar Functions. In order to provide a prediction in real-time, the model usually receives a single data point from the caller, and is expected to provide an accurate prediction within a few milliseconds. 

Throughout this talk, I will demonstrate the steps required to deploy a fully-trained ML that predicts the delivery time for a food delivery service based upon real-time traffic information, the customer's location, and the restaurant that will be fulfilling the order. 

- PDT
[#VIRTUAL] OPEN TALK (AI): How To Build An AI Based Knowledge Graph for Customers in Fintech
Gautam Gupta
Gautam Gupta
Intuit, Technology leader

In this session, we’d go through our journey to build an AI based Customer Knowledge graph. We’d share the insights & knowhow required to create this scalable & polyglot data platform. Join us to learn the design patterns & best practices that we have developed over time to create an intelligent solution based on AI & Graph technologies for an ever increasing list of product lines and customers. 

- PDT
[#VIRTUAL] OPEN TALK (AI): Patenting Artificial Intelligence– How AI Companies Can Identify and Protect AI Inventions
Steve Bachmann
Steve Bachmann
Bachmann Law Group PC, President, Silicon Valley Patent Attorney

Artificial intelligence is becoming one of the most widespread and useful technologies in use today. From data collection to model training, language processing to predictive models, deep networks to AI frameworks, there are many categories and implementations of AI, all with protectable features and important business applications. Protecting cutting edge AI technology helps companies achieve business goals and support their AI innovation.
This presentation will identify key strategies to identify which aspects of AI are patentable and which aspects are not. The discussed strategies will be supplemented with practical real-world examples of patenting different areas of the AI process, from data collection to model training and model implementation to output applications, as well as distinct types of AI systems.
Attendees will also learn about AI patent trends and the most common use cases in which different AI companies build valuable patent portfolios around their AI technology. 

- PDT
[#VIRTUAL] OPEN TALK (AI): Scalable, Explainable and Unsupervised Anomaly Detection for Telecom
Ivan Caramello de Andrade
Ivan Caramello de Andrade
Encora Brazil Division, Innovation Leader and Tech Lead

In developing and implementing a telecommunications network, one of the most oppressive challenges that these companies deal with are anomalies that occur within the network showing that something strange (usually an attack, a fraud or an error) is happening. Detecting these anomalies is a challenge because they may appear in different places and formats and require the observation of multiple metrics over hundreds of thousands of events to tell regular behaviors from anomalous ones. Ivan Carmello De Andrade, would like to explain how detecting these anomalies with higher accuracy may be possible with the technology and machine learning capabilities of today.

In his technical session, Ivan will explain how he and his team were able to customize and adapt a Robust Random Cut Forest model to identify and explain anomalies in an unsupervised and scalable way. He and his team will explain the process behind creating this solution as well as the challenges they overcame in development, such as extracting behaviors from individual events. He will also explain the benefit of this model to the user which include:

• The user does not need to understand which behaviors are regular or anomalous nor which features are relevant to describe and identify them
• The model provides accountability, because the user can identify and understand which factors lead to an event being identified as an anomaly
• Scalability in general, the model can be implemented on many different scales with a highly distributable structure and configurable levels of detail 

- PDT
[#VIRTUAL] OPEN TALK (AI): Pushing Deepfakes to the Limit - Fake Video Calls with AI
Thomas Endres
Thomas Endres
TNG Technology Consulting GmbH, Partner
Martin Förtsch
Martin Förtsch
TNG Technology Consulting GmbH, Principal Consultant
Jonas Mayer
Jonas Mayer
TNG Technology Consulting GmbH, Senior Consultant

Today's real-time Deepfake technology makes it possible to create indistinguishable doppelgängers of a person and let them participate in video calls. Since 2019, the TNG Innovation Hacking Team has intensively researched and continuously developed the AI around real-time Deepfakes. The final result and the individual steps towards photorealism will be presented in this talk.

Since its first appearance in 2017, Deepfakes have evolved enormously from an AI gimmick to a powerful tool. Meanwhile different media outlets such as "Leschs Kosmos", Galileo and other television formats have been using TNG Deepfakes.

In this talk we will show the different evolutionary steps of the Deepfake technology, starting with the first Deepfakes and ending with real-time Deepfakes of the entire head in high resolution. Several live demos will shed light on individual components of the software. In particular, we focus on various new technologies to improve Deepfake generation, such as Tensorflow 2 and MediaPipe, and the differences in comparison to our previous implementations. 

- PDT
[#VIRTUAL] OPEN TALK (AI): Democratizing Deep Learning with Vector Similarity Search
Nava Levy
Nava Levy
Redis, AI/ML Developer Advocate

Deep learning is responsible for most of the breakthroughs we have seen in AI/ML in recent years, yet most companies' models in production use classic or traditional ML. In this talk we will explore how deep learning is being democratized today, thanks to the rising use and availability of vector embeddings from giant pre-trained neural networks. We will see how these embeddings can be combined together with vector similarity search to address different use cases covering any modality and applied to any type of object. Finally, we will discuss the many opportunities this presents as well as the tools that are required to successfully deploy these applications into production. 

Thursday, November 3, 2022

- PDT
[#VIRTUAL] KEYNOTE (AI): Indico Data - Unstructured Data: Challenge and Opportunity for the AI Developer
Christopher M. Wells, Ph. D.
Christopher M. Wells, Ph. D.
Indico Data, VP of Research and Development

Unstructured Data represents a massive and little explored frontier for both the enterprise and the enterprise technology professional. The dizzying proliferation of tools for programatically working with documents, audio, images and video (as well as the corresponding hype) can be overwhelming. This session will provide a practical framework for breaking down the analysis and automation of unstructured data stores and flows, as well as a survey of success stories. 

- PDT
[#VIRTUAL] OPEN TALK (AI): Deep Dive on Creating a Photorealistic Talking Avatar
Sebastiano Galazzo
Sebastiano Galazzo
Synapsia.ai, Artificial intelligence researcher

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 

- PDT
[#VIRTUAL] PRO TALK (AI): Avoid Mistakes Building AI Products
Karol Przystalski
Karol Przystalski
Codete, CTO

Based on Gartner's research, 85% of AI projects fail. In this talk, we show the most typical mistakes made by the managers, developers, and data scientists that might make the product fail. We base on ten case studies of products that failed and explain the reasons for each fail. On the other hand, we show how to avoid such mistakes by introducing a few lifecycle changes that make an AI product more probable to succeed. 

- PDT
[#VIRTUAL] FEATURED TALK (AI): Circumventing Scripting: Automating Conversation Design
Keren Douek
Keren Douek
BOTS, Founder and Chief Conversation Designer

Whether building a chatbot with or without code, the scripting process remains a behemoth task. We're looking at all the ways Conversational Design can be automated, to make building a chatbot script less burdensome and open up the field to creative users who can help exponentially expand chatbot use cases. At BOTS, we strive to get creative users building chatbots and A.I. solutions regardless of background. This year, we launched a STEM version in the schools where students in K-5 built their own chatbots to support their lesson and learn about A.I. 

- PDT
[#VIRTUAL] PRO TALK (AI): Leveraging Automated Machine Learning to Enable Anyone to Develop Machine Learning Solutions
Hugo Barona
Hugo Barona
Ergo, Cloud Solution Architect

Nowadays, several business owners know that leveraging Artificial Intelligence capabilities, on their systems and applications, can enable their businesses to achieve better results. But building Artificial Intelligence solutions may be a time-consuming and complex process, so consequently, some of these people give up of building such solutions, since they or their team do not have the expertise and capacity required, or sometimes they end-up paying to third-party companies to build these solutions and as a consequence, they end-up doing a significant investment on building these solutions. Azure Automated Machile Learning is the solution to enable anyone to build the Artificial Intelligence and Machine Learning solutions at low cost and with the best quality possible. 

- PDT
[#VIRTUAL] PRO TALK (AI): Physics-Based Graph Neural Networks Enable Composable, Strongly Typed Neural Networks
Troy Harvey
Troy Harvey
PassiveLogic, Co-founder, CEO, and Product Architect

PassiveLogic’s (www.passivelogic.com) platform for generalized autonomy utilizing Deep Digital Twins is built on systems-level control theory. The platform is generalized because it can be used to control any kind of system. At its core, this type of platform works on the sensor-fusion and control-fusion of digital models. In these Deep Digital Twin models, the digital twin literally is the AI structure. Each digital twin utilizes the fundamentals of physics to model a single component or piece of equipment. When multiple digital twins are linked to each other in a graph neural network, they form a system description. Because their physics are integral to the models themselves, these graph-based system descriptions model not only the real complexities of systems but also their emergent behavior and the system semantics.
Deep physics networks are structured similar to neural networks, but unlike the homogeneous activation functions of neural nets, each neuron comprises unique physical equations representing a function in a thermodynamic system. The Deep Physics approach is built on heterogeneous neural nets that are composable, have physics guarantees, allow users to define their own systems, learn unsupervised, and generate a physics description of a system. Being so principled, it is also necessarily more constrained, meaning the physics-based graph neural networks can be used to predict future system behavior.
The physics-based graph neural network provides a systems-level intelligence as it understands the interconnectivity of components in a system. As such, it can automatically infer behavior and introspect results, even where sensors do not exist. Using this inference ability, an autonomous control platform built on Deep Digital Twins can provide self-commissioning, automate point-mapping, validate installation, and provide continuous system measurement and verification against its original design. Real-time system operational data can be brought into the model for real-time machine learning so that the model can adapt for improved accuracy of predicting the system behavior.
In this talk, Troy Harvey, CEO at PassiveLogic, will describe Deep Digital Twin AI structures and the applications for generalized autonomy. 

- PDT
[#VIRTUAL] OPEN TALK (AI): Operationalizing AI with a Shift from Research to Product Orientation
Yotam Oren
Yotam Oren
Mona, CEO & Cofounder

Many AI programs fail to deliver sustained value despite great research, due to insufficient operational tools, processes and practices. These days, more and more data science teams are going through a major shift, from research orientation, to product orientation. Key factors to successfully transition to a product-oriented approach to AI include empowering data scientists to take end to end accountability for model performance, and going beyond the model - gaining a granular understanding of the behavior of the entire AI-driven process. In this talk, Yotam will discuss the importance of empowering data science teams to successfully make the transition from research oriented to product oriented. 

- PDT
[#VIRTUAL] OPEN TALK (AI): Level Up Your Data Lake - to ML and Beyond
Oz Katz
Oz Katz
Treeverse, CTO & Co-Founder

A data lake is primarily two things: an object store and the objects being stored. Even with the most basic setup, data lakes are capable of supporting BI, Machine Learning, and operational analytics use cases. This flexibility speaks to the strength of object stores, particularly their flexibility in integrating with a diverse set of data processing engines.

As data lakes exploded in adoption, a number of improvements were made to the first architectures. The first and most obvious improvement was to file formats, which led to the development of analytics-optimized formats like parquet, and eventually modern table formats.

An even newer improvement has been the emergence of data source control tools that bring new levels of manageability across an entire lake! In this talk, we'll cover how to incorporate these technologies into your data lake, and how they simplify workflows critical to ML experimentation, deployment of datasets, and more! 

- PDT
[#VIRTUAL] OPEN TALK (AI): Reducing Latency and Resource Consumption for Offline Feature Generation
Dhaval Patel
Dhaval Patel
Netflix, Machine Learning Infrastructure

Personalization is one of the key pillars of Netflix as it enables each member to experience the vast collection of content tailored to their interests. Our personalization system is powered by various machine learning models. We constantly innovate by adding new features to our personalization models and running A/B tests to improve recommendations for our members. We also continue to see that providing larger training sets to our models helps make better predictions. Our ML fact store has enabled us to provide larger training sets where the training set spans over a long time window. While a great success, the ML fact store architecture has its limitations. For example, features computed while generating recommendations must be recomputed by offline feature generation pipelines. This talk is about those limitations and how we enhanced our architecture to run optimized offline feature generation pipelines. 

- PDT
[#VIRTUAL] OPEN TALK (AI): Bringing Life and Motion to AI Explainability
Joao Nogueira
Joao Nogueira
Optum, Senior AI Engineer
Pietro Mascolo
Pietro Mascolo
Optum Ireland, Data Scientist

SHAP is a great tool to help developers and users understand black box models. To push it to the next level, we will show how to leverage on Dash, SHAP, gifs, and auto-encoders to generate interactive dashboards with animations and visual representations to understand how different AI models learn and change their minds while progressively trained with growing amounts of data.

Animations will help developers understand how frequently AI models tweak their population and local importance factors during training and how they compare across competing AI models, adding an extra layer to AI safety. Auto-encoders and LSTM will be used to generate 2-dimensional embedding representations of explainability paths at individual level, allowing developers to interactively detect algorithm decision making similarity across time and visually debug mislabeled AI predictions at each point in time.

We will show this application in the context of Chronic Kidney Disease prediction and broader Healthcare AI.