Ceta

Level 2 - Certified Enterprise Transformation Analyst (CETA)

CETA Emblem

Course Curriculum

๐—–๐—˜๐—ง๐—” ๐— ๐—”๐—œ๐—กย โ€”ย ๐—”๐—ฑ๐˜ƒ๐—ฎ๐—ป๐—ฐ๐—ฒ๐—ฑ ๐—”๐—œ,ย ๐—Ÿ๐—Ÿ๐— ย &ย ๐—”๐—ด๐—ฒ๐—ป๐˜๐—ถ๐—ฐ ๐—ฆ๐˜†๐˜€๐˜๐—ฒ๐—บ๐˜€
40+ Advanced Models, Architectures & Systems

1. ๐—Ÿ๐—ฎ๐—ฟ๐—ด๐—ฒ ๐—Ÿ๐—ฎ๐—ป๐—ด๐˜‚๐—ฎ๐—ด๐—ฒ ๐— ๐—ผ๐—ฑ๐—ฒ๐—น๐˜€ (๐—Ÿ๐—Ÿ๐— ๐˜€)

Foundation models trained on large-scale corpora

1. Text Tokens
2. Transformer
3. Contextual Intelligence
Models: GPT-1 โ†’ GPT-4, BERT, RoBERTa, ALBERT, DistilBERT, T5, XLNet
Focus: Scaling laws, pretraining vs. fine-tuning, enterprise trade-offs

2. ๐—ง๐—ฟ๐—ฎ๐—ป๐˜€๐—ณ๐—ผ๐—ฟ๐—บ๐—ฒ๐—ฟ ๐—”๐—ฟ๐—ฐ๐—ต๐—ถ๐˜๐—ฒ๐—ฐ๐˜๐˜‚๐—ฟ๐—ฒ (๐——๐—ฒ๐—ฒ๐—ฝ ๐——๐—ถ๐˜ƒ๐—ฒ)

Self-attentionโ€“based sequence modeling

1. Embeddings
2. Self-Attention
3. Multi-Head Attention
4. Output
Components: Qโ€“Kโ€“V, Positional Encoding, Residuals, Layer Norm
Math: Softmax attention, Dot-product scaling, Dimensionality management

3. ๐—ง๐—ฟ๐—ฎ๐—ป๐˜€๐—ณ๐—ผ๐—ฟ๐—บ๐—ฒ๐—ฟ ๐—ข๐—ฝ๐˜๐—ถ๐—บ๐—ถ๐˜‡๐—ฎ๐˜๐—ถ๐—ผ๐—ป & ๐—ฆ๐—ฐ๐—ฎ๐—น๐—ถ๐—ป๐—ด

Reduce cost, increase context, improve efficiency

1. Long Input
2. Optimized Attention
3. Scalable Output
Techniques: Sparse Attention, FlashAttention, Longformer, BigBird, Reformer
Focus: Computational complexity $O(n^2)$ vs linear scaling, memory bottlenecks

4. ๐— ๐—ถ๐˜…๐˜๐˜‚๐—ฟ๐—ฒ ๐—ผ๐—ณ ๐—˜๐˜…๐—ฝ๐—ฒ๐—ฟ๐˜๐˜€ (๐— ๐—ผ๐—˜)

Sparse activation of expert sub-models

1. Input Embedding
2. Router Mechanism
3. Expert Selection
4. Weighted Output
Models: Switch Transformer, GLaM, Mixtral (SMoE), DeepSeek-V3
Focus: Scaling parameters without increasing FLOPs, conditional computation

5. ๐—ฆ๐˜๐—ฎ๐˜๐—ฒ ๐—ฆ๐—ฝ๐—ฎ๐—ฐ๐—ฒ ๐— ๐—ผ๐—ฑ๐—ฒ๐—น๐˜€ (๐—ฆ๐—ฆ๐— ๐˜€)

Linear-time sequence modeling

1. Sequence
2. State Update
3. Output
Models: SSM, Mamba
Advantages: Long-context, low-memory, edge-friendly

6. ๐—ฅ๐—ฒ๐˜๐—ฟ๐—ถ๐—ฒ๐˜ƒ๐—ฎ๐—น-๐—”๐˜‚๐—ด๐—บ๐—ฒ๐—ป๐˜๐—ฒ๐—ฑ ๐—š๐—ฒ๐—ป๐—ฒ๐—ฟ๐—ฎ๐˜๐—ถ๐—ผ๐—ป (๐—ฅ๐—”๐—š)

Ground LLM outputs using external knowledge

1. Query
2. Retriever
3. Context
4. LLM
5. Answer
Architectures: Basic RAG, Hybrid RAG, Multi-Hop RAG
Tools: FAISS, Pinecone, Weaviate

7. ๐—”๐—ด๐—ฒ๐—ป๐˜๐—ถ๐—ฐ ๐—ฅ๐—”๐—š

RAG + reasoning + action

1. Goal
2. Retrieve
3. Reason
4. Act
5. Verify
Capabilities: Self-Reflection, Multi-Step Planning, Tool Use
Used in: legal AI, finance, enterprise automation

8. ๐—”๐—ด๐—ฒ๐—ป๐˜๐—ถ๐—ฐ ๐—”๐—œ ๐—ฃ๐—ฎ๐˜๐˜๐—ฒ๐—ฟ๐—ป๐˜€

Control how models think and speak

1. Text
2. Tokens
3. Probabilities
4. Output
Tokenizers: BPE, WordPiece, Unigram
Decoding: Greedy, Beam, Top-K, Top-P, Temperature

9. ๐—”๐˜‚๐˜๐—ผ๐—ป๐—ผ๐—บ๐—ผ๐˜‚๐˜€ ๐—”๐—ด๐—ฒ๐—ป๐˜๐˜€

Systems that execute multi-step goals

1. Objective
2. Planning
3. Execution
4. Monitoring
Autonomy levels: L1 (Assistive), L2 (Bounded), L3 (Limited Autonomy)
Safeguards: caps, approvals, audit trails

10. ๐— ๐˜‚๐—น๐˜๐—ถ-๐— ๐—ผ๐—ฑ๐—ฎ๐—น ๐—”๐—œ

Joint reasoning across data types

1. Text + Image + Audio
2. Unified Model
3. Output
Models: CLIP, DALLยทE, GPT-V, Whisper, Wav2Vec 2.0, VideoGPT

11. ๐—ฉ๐—ถ๐˜€๐—ถ๐—ผ๐—ป ๐—ง๐—ฟ๐—ฎ๐—ป๐˜€๐—ณ๐—ผ๐—ฟ๐—บ๐—ฒ๐—ฟ๐˜€ (๐—ฉ๐—ถ๐—ง)

Transformers beyond language

1. Image
2. Patch Embedding
3. Self-Attention
4. Prediction
Models: ViT, Hybrid CNN-ViT

12. ๐—œ๐—ป๐˜€๐˜๐—ฟ๐˜‚๐—ฐ๐˜๐—ถ๐—ผ๐—ป ๐—ง๐˜‚๐—ป๐—ถ๐—ป๐—ด & ๐—ฅ๐—Ÿ๐—›๐—™

Align models with human intent

1. Prompt
2. Reward Model
3. Policy Update
Techniques: Supervised Fine-Tuning, RLHF, FLAN-T5, InstructGPT

13. ๐—ฃ๐—ฎ๐—ฟ๐—ฎ๐—บ๐—ฒ๐˜๐—ฒ๐—ฟ-๐—˜๐—ณ๐—ณ๐—ถ๐—ฐ๐—ถ๐—ฒ๐—ป๐˜ ๐—™๐—ถ๐—ป๐—ฒ-๐—ง๐˜‚๐—ป๐—ถ๐—ป๐—ด (๐—ฃ๐—˜๐—™๐—ง)

Adapt models without retraining everything

1. Base Model
2. Adapter
3. Task-Specific Output
Methods: LoRA, Prefix-Tuning, Adapters, BitFit

14. ๐—ง๐—ผ๐—ธ๐—ฒ๐—ป๐—ถ๐˜‡๐—ฎ๐˜๐—ถ๐—ผ๐—ป & ๐——๐—ฒ๐—ฐ๐—ผ๐—ฑ๐—ถ๐—ป๐—ดโ€‹

Control how models think and speak

1. Text
2. Tokens
3. Probabilities
4. Output
Tokenizers: BPE, WordPiece, Unigram
Decoding: Greedy, Beam, Top-K, Top-P, Temperature

15. ๐—š๐—ฒ๐—ป๐—”๐—œ ๐—ข๐—ฝ๐˜€ (๐—ฃ๐—ฟ๐—ผ๐—ฑ๐˜‚๐—ฐ๐˜๐—ถ๐—ผ๐—ป ๐—”๐—œ)โ€‹

Design grammar for autonomy

1. Task
2. Agent Pattern
3. Controlled Execution
Patterns: Plan-Act-Reflect, Supervisor-Worker, Reflexion Loop, Task Graphs
Scroll to Top