5 OCI AI Portfolio
AI Infrastructure:
a. Oracle AI Stack: Components not specified. b. AI Infrastructure Components: GPU, Networking, Super Clusters, Storage. c. GPU Architecture: i. Capabilities: Parallel computing, rapid computations, processing large datasets. d. OCI GPU Instances: Ideal for model training and interference computation.
ML Services Overview:
a. Oracle AI Stack: Components not specified. b. Oracle Cloud Infrastructure Data Science: i. Definition: Cloud service for data scientists throughout ML lifecycle. ii. Core Principles: Accelerated, Collaborative, Enterprise-Grade. iii. What, Whom, Where, How: Builds, serves data scientists, Jupyter Lab, Model Catalog. c. Data Science Features and Terminology: i. Components: Projects, Notebook Sessions, Conda Environments, ADS SDK, Models, Model Catalog, Model Deployment, Jobs.
First Principles:
a. RDMA: i. Definition Remote Direct Memory Access. b. Clos Fabric: i. Definition: Multistage circuit-switching network. Visioned by Charles Clos in 1950s. c. Cable Distance: Considerations for higher maintenance; account for worst-case scenarios.
Overview of AI Services:
i. AI Service Types: Language, Vision, Speech, Document Understanding, Anomaly Detection, Digital Assistant.
AI for the Enterprise:
i. Components: SaaS Apps, AI Services, Infrastructure, Data. ii. Oracle Focus: Bringing AI to every layer of the enterprise stack. iii. Approach: Extensive investment from infrastructure to SaaS apps. iv. Recent Steps: Emphasis on Generative AI and massive scale models. v. Oracle AI Stack: Components not specified.
Human Ethics and Fundamental Rights:
i. Human Ethics: Respect for human dignity, freedom, democracy, justice, equality. ii. AI Ethics: Respect for autonomy, prevention of harm, fairness, explicability. iii. Responsible AI Requirements: Human-centric, technical robustness, privacy, transparency, diversity, accountability.
Ways to Access Oracle Cloud AI Services:
i. Methods: OCI Console, Rest API, Language SDKs, CLI.
Language Overview:
i. Pretrained Models: Language Detection, Sentiment Analysis, Key Phrase Extraction. ii. Custom Models: Named Entity Recognition, Text Classification. iii. Text Translation: Neural machine translation for various languages. iv. Vision: Pretrained and custom models for image analysis. v. Speech: Not detailed. vi. Document Understanding: OCR, Text Extraction, Key-Value Extraction, Table Extraction, Document Classification. vii. Anomaly Detection: Multitenant service analyzing time series data. viii. Digital Assistant: Interacts, lists capabilities, routes request, handles disambiguation.
Healthcare AI Challenges:
Balancing AI benefits with concerns about fairness, transparency, and regular evaluation.
Trustworthy AI:
Driven by ethical principles.
Guiding Principles for Trustworthy AI:
Follow applicable laws, be ethical, be robust.
Responsible AI Cycle and Roles:
Implementing and managing AI according to responsible AI requirements. Roles: Developers, Deployers, End Users.
AI Needs to be Lawful:
Operates within legal frameworks.
Responsible AI requirements
Set up governance to be put into place. Develop policies and procedures. Ensure Compliance
