NVIDIA · NCA-GENM
The compute-efficiency thread
Getting more accuracy per unit of compute. Opens in M1 with training-stability levers (normalization, LR warmup, gradient clipping), and concentrates in M5's mixed precision, quantization, and pruning before M6 puts TensorRT and Triton underneath the served pipeline.
NCAM-T3 · 24 lessons across 3 modules
- M1M1-01Machine learning fundamentals: learning paradigms, feature engineering, and cross-validation
- M1M1-02Overfitting, underfitting, and the bias-variance tradeoff
- M1M1-03Model comparison metrics: accuracy, precision, recall, F1, ROC-AUC, MAE, MSE, R²
- M1M1-04Deep learning frameworks: TensorFlow, PyTorch, and Keras
- M1M1-05Neural network basics: neurons, activation functions, and the training loop
- M1M1-06Convolutions and the building blocks of vision models
- M1M1-07Nonsequential networks and residual connections
- M1M1-08Multimodal loss functions: cross-entropy, contrastive, reconstruction, adversarial, and composite
- M1M1-09Training stability in multimodal settings: normalization, LR warmup, loss weighting, gradient clipping
- M1M1-10Multimodal transfer learning: pretrained encoders, full fine-tuning vs. parameter-efficient adaptation
- M1M1-11Model fusion and orchestration: early, intermediate, and late fusion; modality vs. agent orchestration
- M1M1-12Prompt engineering fundamentals and emerging multimodal trends
- M5M5-01Mixed-precision training: FP16, FP32, loss scaling, and Tensor Cores
- M5M5-02Quantization: PTQ vs. QAT
- M5M5-03Neural network pruning: structured vs. unstructured
- M5M5-04Hyperparameter tuning: grid, random, and Bayesian search
- M5M5-05Transfer learning for efficiency
- M5M5-06Energy efficiency and inference optimization with TensorRT and Triton
- M6M6-01U-Net architecture: encoder-decoder structure and skip connections
- M6M6-02The U-Net as diffusion denoising backbone and as an autoencoder
- M6M6-03CLIP plus diffusion: building a text-to-image pipeline
- M6M6-04NVIDIA SDKs: NeMo, Riva, Triton, ACE, cuDNN, and AI Blueprints/VIA
- M6M6-05Prompt engineering for generative systems, and software quality practices
- M6M6-06Putting it together: a text-to-image service end to end
Part of the throughlines running across the NCA-GENM prep course.