Untitled Quiz

Podcast

Play an AI-generated podcast conversation about this lesson

Questions and Answers

What does Deep Learning utilize to process information?

Static decision trees

Traditional algorithmic methods

Simple linear equations

Complex neural networks (correct)

How does Natural Language Processing (NLP) enhance user experience?

By simplifying coding languages

By generating human-like speech or text (correct)

By optimizing database queries

By increasing image resolution

Which statement best describes Machine Learning?

It allows machines to mimic human learning and execution. (correct)

It only focuses on mathematical problem-solving.

It requires manual oversight for learning.

It can only analyze structured data types.

What is a key application of Computer Vision?

Interpreting digital images and videos Signup and view all the answers

What characterizes Reinforcement Learning in Machine Learning?

It learns through trial and error. Signup and view all the answers

How do Recommendation Engines function?

By predicting user preferences based on historical data Signup and view all the answers

What is the primary purpose of using AI in business outcomes?

To drive insights and support decision-making Signup and view all the answers

What aspect of AI applications is linked to Intelligent Robotics?

Replicating human actions and tasks Signup and view all the answers

What is a major benefit of neural networks when it comes to learning from new data?

They can adapt and evolve with new data. Signup and view all the answers

Which aspect of evolutionary AI allows it to solve complex optimization tasks?

Employing mutation and crossover between solutions. Signup and view all the answers

What is a limitation of neural networks in their decision-making process?

Their decision-making process is opaque and complex. Signup and view all the answers

In what application can neural networks be notably effective?

Image recognition tasks. Signup and view all the answers

Which feature of evolutionary AI enhances its problem-solving capability in uncertain environments?

Resilience through iterative improvement. Signup and view all the answers

What is the primary mechanism through which neural networks learn?

Adjusting the strength of connections between neurons. Signup and view all the answers

What is a common challenge when using neural networks with respect to training data?

Lack of generalization to new data. Signup and view all the answers

What type of tasks are powered by connectionist AI in natural language processing?

Sentiment analysis and language translation. Signup and view all the answers

What is a key characteristic of Symbolic AI?

Processes information through explicit rules and logical operations. Signup and view all the answers

What limitation does Symbolic AI face when dealing with real-world applications?

Struggles with incomplete knowledge and adaptability. Signup and view all the answers

Which of the following is an example of a task that Symbolic AI is well-suited for?

Mimicking human expertise in medical diagnosis. Signup and view all the answers

What aspect of Symbolic AI provides transparency during its decision-making process?

Explicit rules governing symbol manipulation. Signup and view all the answers

In Natural Language Processing, which capability enhances human-computer interaction?

Understanding and generating human language. Signup and view all the answers

What is a characteristic of Reinforcement Learning?

It uses a trial-and-error approach to maximize rewards. Signup and view all the answers

Which of the following correctly describes the scalability issue in Symbolic AI?

The expense in computational power as symbols and rules increase. Signup and view all the answers

Which approach to AI typically lacks adaptability to new situations not covered in its existing knowledge base?

Symbolic AI. Signup and view all the answers

Study Notes

SBG Data & AI Value Tower

Al Advisory Training Days on November 21st & 22nd, 2024
Event theme: "Navigate the Data Jungle with Ease - SAP Data & AI"

Agenda Day 1

Introduction to Al (08:45 - 09:30) with Dauren Eshenkulov
Introduction to GenAl (09:30 - 10:00) with Dauren Eshenkulov
Associate with Prompts (10:30 - 11:45) with Thuy Thu Do
Winning Use Case GenAl Hackathon (11:45 - 12:15) with Marina Alej. Gonz. Petit
Lunch Break (12:15 - 13:30)
Use Cases from GenAl Hackathon (13:30 - 15:00) with Amit Bi Kumar
Break (15:00 - 15:15)
Al in Cybersecurity (15:15 - 15:45) with Thuy Thu Do
Dangers of Al (15:45 - 16:45) with Deniz Löwner
Q&A/Closing (16:45 - 17:15) with Amit & Deniz

Our trainers

Amit Bi Kumar
Marina Alejandra Gonzalez Petit
Vivek Krishnan Kadasamy
Dauren Eshenkulov
Vera Khobotova
Deniz Löwner
Thu Thuy Do
Chitransh Asthana
Valerie Minderjahn
Dylan Lott
Diana Torre

Data & Al – value tower in ASG

Cooperation with Data & Al domain and data factories, e.g., ATCI, India, Philippines, Riga
Opportunity & Pipeline
People
Recruiting
Operations
Data strategy & architecture
Data & Al Phase II (to be added)

Al to advance SAP Data migration

Purpose: Identify Al use cases that can advance data migration. Conduct use case development. Partner with clients for pilot development. Al upskilling via training. Understand data migration tools for Al capabilities. Jump start training enhancement with Al topics
Main Responsibilities: Al use case building for SAP data migration. Al Upskilling. Pilot development /implementations. Participation on Hackathon or tech meetups. Integration within and outside the value tower. Collaboration with data migration, data governance, SNP, data quality
Initially assigned team members: Amit Kumar (lead), Korn, Sascha, Kantawala, Anand, Gonzalez Petit, Marina, etc.

Your time for questions

Join at menti.com
Use code 2524 9982

Introduction to Al

Presenter: Dauren Eshenkulov

Welcome to the "Age of Al"

1950s – early 2000s: Machines replicate, better than humans. Research on neural networks
Early 2000s – late 2010s: Machines learn, better than humans. Deep learning revolution (NLP). Al beats humans at Go.
Late 2010s – today: Machines create, better than humans. Transformer architecture. Al achieves supremacy in language. Compute capacity trends from TFLOPS to billions. Data availabilty trends

Definition of Al

Self-learning technologies that sense, comprehend, and act to drive new business outcomes.
Machine Learning: Machines mimic human learning and task execution
Deep Learning: Uses neural networks (thousands or millions of interconnected nodes)
Reinforcement Learning: Learning through trial and error

Types of Al

Narrow Al (Weak Al): Systems trained for specific tasks (e.g., virtual assistants, image recognition)
General Al (Strong Al): Systems with human-level intelligence across various tasks
Superintelligent Al: Hypothetical Al systems surpassing human intelligence

Approaches to Al - Symbolic Al

Uses explicit rules and logical operations to process information and make decisions.
Interpretability, knowledge representation, and flexibility.
Limitations: Incomplete knowledge, scalability, and limited adaptability.
Examples: Expert systems, natural language processing and applications

Approaches to Al – Connectionist Al

Inspired by the human brain structure and function.
Learning through adjusting connection strengths between artificial neurons.
Adaptable to new data, resilient to damage.
Limitations: Performance reliant on data quality, opaque decision-making, and risk of overfitting to training data.

Approaches to Al – Evolutionary Al

Draws inspiration from natural selection and evolution.
Using algorithms that mimic biological evolution to solve problems.
Suitable for tasks requiring optimal solutions rather than perfect ones.
Limitations: high computational overhead, sensitive to parameter settings, complex optimization problems.

Introduction to Gen Al

Presenter: Dauren Eshenkulov

GenAl is much more than just a trend: it's a reinvention

Every 14 years, there are "Cambrian explosions" of technological advancement (internet, smartphones). Now, that is happening with GenAl in 2022.

Why Gen Al is so relevant?

Large market size ($641.3 billion valued by 2028).
Expected addition to global economy by 2030 ($15.7 trillion).
38.1 percent compound annual growth rate for global enterprise adoption (2022-2030)

Generative Al vs Traditional Al

Generative Al: Creates data, audio, media. Uses Large Language Models (LLMs) to predict.
Traditional Al: Uses specific instructions to follow.

GenAl has the capability to learn and reapply patterns of data for a wide range of applications

Foundation model trained once, can generate output from various input types (text, image, video)

...has the capability to impact the whole organization

Product Differentiation and Innovation: 3D product designs, creative sessions, personalized offerings
Operations and Supply Chain: Real-time analytics, avoid bottlenecks, and predictive analytics with real-time monitoring
Sales and Marketing: Targeted advertising, personalized content, consumer segmentation and insight
IT Development: Testing and coding assistance, streamline workflows, generate synthetic data
Consumer Experience: Personalized purchase experiences, chatbot service, inclusive product offering
Support Functions: Develop training content, self-serve systems, automate tasks

Gen Al augments the enterprise Al continuum

Diagnostic: Analyze, scenario, segment.
Predictive: Pattern, forecast, model.
Generative: Create, automate, protect

The Main Ingredients

Deep (Memory) Neural Network: Layers of interconnected nodes (neurons) mimic the brain.
GPU Architectures: Processing power critical for training and running Al models.
Data: Massive datasets are essential for model training, ranging from Tbytes to Pbytes

Generative Pre-trained Transformer Training and Inference Procedure

Supervised pre-training to fine-tune various models.

Today's Gen Al investments span three types of use cases that vary based on domain specificity

Cross-industry productivity: Large models (e.g., productivity copilot, etc.).
Domain-specific tasks: Focused models for specific issues or use cases.
Complex industry problem solving: Smaller, fine-tuned models (e.g., NOC operations optimization)

New ways of working: unlocked creativity

By 2025, marketing organizations will shift 75% of their staff's operations to strategic tasks using AI

New ways of working: highly personalized experiences

Customers are treated individually based on their characteristics.
Data-driven actionable insights from multi sources are considered for decision-making
Highly personalized experiences are tailored to needs and interests.

Gen Al is growing -every industry is experimenting

50% of executives will make significant investments in Gen Al in 2024.
Fewer than 10% are reaching value with Gen Al.

Gen Al Ecosystem Landscape

Applications: Out-of-the-box offerings; end to end solutions.
Foundation Models: Generate, image, video.
Data & Platforms: Store and manage data.
Cloud & Infrastructure: Compute and infrastructure

Associate with Prompts

Presenter: Thu Thuy Do.
Agenda: What is prompt, what is prompt engineering, different types of prompts, prompting techniques, elements of good prompt

What is prompt?

Text input (question, command, etc.) to trigger an Al model's response
Prompting is the process of directing an Al to perform a task

Examples of prompt

Quality of input impacts the quality of output.
Simple changes in prompts can dramatically change results.

What is prompt engineering?

The process of crafting and optimizing prompts to get desired outputs from an Al model. Two main purposes of prompt engineering are discovery and adaptation.

Types of prompt

Question & Answer, Text Completion, Language Translation, Summarization, Creative Writing, Code Generation, Extraction

Components of a good prompt

Task, Context, Format, Persona, Tone, Examples

Task

Direct description of the desired outcome

Context

Background information needed to understand the task requirements

Format

Structure and style of the prompt

Tone

The voice/emotion you use to communicate with the model (e.g, professional, friendly, approachable)

AI at SAP

Presenter: Vera Khobotova.
Generative Al toolset for developers to customize and use already in SAP Business Technology Platform.

SAP BTP Solutions with Built-in Al

SAP Build Code: Generative Al-based code development (Joule copilot). Tailored for SAP development and enhanced fusion.
SAP HANA Cloud: Limitless database for any workload, intelligent data apps (data analytics).
SAP Build Process Automation: Data extraction, structured documents, data enrichment
SAP Analytics Cloud: Generate Al business plans, automatic reporting, elevation of BI capabilities

Al built for business

ERP and Finance
Supply Chain, Procurement
Human Resources
Sales & Marketing
IT and Platform
Industries

How did Accenture Transform Lead-to-Cash Processes with Ready-to-Use AI?

Increase in accuracy (54%).
Time savings (reduced months spent).
SAP Cash application software automates clearing, provides machine learning-generated invoice matching model

Accenture HR Audit and Compliance as-a-service

On-going data quality assurance
Data analysis, gaining insight into patterns and dependencies in current data
Data correction based on machine learning algorithm propositions

Hands-on: Alejandria

Presenter: Marina Gonzalez Petit.

Extending applications with GenAl

Presenter: Vivek Krishnan Kandasamy

Gen Wizard

Presenter: Amit Bi Kumar.

GenAI Development Building Use Cases

Presenter: Chitransh Asthana.

GenAl Quiz

Presenter: Diana Torre.

Q&A

Presenter: Amit Bi Kumar

Al Trends

Presenter: Dylan Lott.
Multimodal Al: Processes diverse data types—text, images, and video.

Agentic Al

Presenter: Peter Norvig.

Edge Al

Al on localized devices enabling real-time processing.
Reduced latency, bandwidth usage, energy efficiency, and more.

Retrieval-Augmented Generation (RAG)

Links generative Al services with information retrieval from external sources.

Customized Enterprise Gen AI Tools

Tools based on narrower models for specific industries
Examples: Finance, legal.
Precision, relevance, greater efficiency

TQ – Technology Quotient

Presenter: Amit Bi Kumar.

Al Foundation on SAP BTP

Artificial intelligence enables computers and machines to simulate human intelligence.

Latest state of ChatGPT

Chat with images, chat with voice, create new images, code interpreter, web browsing, individual creation (ChatGPT Plus)

Al in Cybersecurity

Presenter: Thu Thuy Do.

Traditional Zero Trust vs Self-Learning Al

Traditional Zero Trust: Predefined, static rules. Slower, and more manual.
Self-Learning Al: Adaptable, Real-time, continuous learning from live data.

Applications of Al in Cybersecurity

Password Protection
Phishing Detection
Vulnerability Management
Network Security
Behavioral Analytics

Al-Powered Cybersecurity Tool - Darktrace

Enterprise Immune System
Darktrace Antigena
Cyber Al Analyst

Dangers of Al

Presenter: Deniz Löwner.
Potential dangers, such as deepfakes, biases are discussed.

Ethical Considerations

Algorithmic bias in AI algorithms and datasets
Privacy and data security issues
Accountability and transparency in AI systems

Economic Impacts

Job displacement and skill shift
Economic productivity and innovation
Income inequality and resilience
Policy implications to safeguard against potential hazards of disruptive technologies

Sora – A prompt to video model (OpenAl Tool)

Al developed by OpenAl for text-to-video generation.
Takes text prompts to create short videos.
Potential impact on creative industries and creation of deepfakes.

Studying That Suits You

Use AI to generate personalized quizzes and flashcards to suit your learning preferences.