NCA-MMGA Multimodal GenAI Associate Study Strategy¶

Study Approach¶

Phase 1: Foundation (1-2 weeks)¶

1. Multimodal concepts - Modalities, architectures, fusion strategies
2. Image generation - Diffusion models, text-to-image, prompting
3. 📖 NVIDIA Build - Try models interactively

Phase 2: Specialized (2-3 weeks)¶

1. Speech AI - ASR, TTS, NVIDIA Riva
2. NVIDIA tools - NIM, NeMo, AI Enterprise
3. Applications - Multimodal RAG, safety, evaluation
4. 📖 Riva Docs
5. 📖 NIM Docs

Phase 3: Exam Prep (1 week)¶

1. Practice scenarios
2. Review key concepts and metrics
3. Focus on NVIDIA-specific tools

Recommended Resources¶

• NVIDIA Build - Model playground
• NVIDIA Riva - Speech AI
• NIM Documentation - Inference microservices
• NVIDIA DLI Courses - Official training
• NVIDIA Developer Blog - Technical articles

Exam Tactics¶

Keywords¶

• "Image understanding" - Vision-Language Model (VLM)
• "Generate images" - Diffusion model
• "Speech" or "voice" - NVIDIA Riva
• "Prompt adherence" - Classifier-free guidance
• "Image quality" - FID score
• "Alignment" - CLIP score
• "Safety" - NeMo Guardrails, content moderation

Common Pitfalls¶

• Diffusion models generate images; VLMs understand images
• CFG scale controls prompt adherence, not quality
• FID is for image quality, CLIP Score is for text-image alignment
• Riva is for speech AI, NIM is for LLM/VLM inference
• Multimodal RAG needs both text and image embeddings

Time Management¶

• 60 minutes for 50-60 questions
• ~1 minute per question
• Flag uncertain questions and return at end