NLP & Computer Vision Engineering

• PyTorch custom vision loaders, Albumentations augmentations, and Transfer Learning with ResNet/EfficientNet
• Object Detection: Anchor boxes, IoU, Non-Maximum Suppression (NMS), and real-time YOLOv8 training
• Semantic & Instance Segmentation: U-Net architectures and Mask R-CNN bounding polygon heads
• Multiple Object Tracking (MOT) using DeepSORT and ByteTrack algorithms for surveillance

• Text Preprocessing: Byte-Pair Encoding (BPE), WordPiece tokenization, and sentence chunking
• Vector Representations: Word2Vec, GloVe, and dense semantic embeddings with Sentence-Transformers
• Recurrent Networks: Bidirectional LSTMs and GRUs for Named Entity Recognition (NER) & POS Tagging
• Sequence-to-Sequence (Seq2Seq) architectures and the Bahdanau additive attention breakthrough

• Scaled Dot-Product Attention, Multi-Head Attention, and Positional Encodings deep dive
• Encoder-Only Models (BERT, RoBERTa) for classification, extraction, and semantic search
• Decoder & Seq2Seq Models (GPT, T5, BART) for abstractive text summarization and machine translation
• Multimodal Vision-Language Models: CLIP embeddings, Zero-Shot image classification, and ViT (Vision Transformers)

• Model Quantization (Post-Training Quantization & Quantization Aware Training), Pruning, and Knowledge Distillation
• Exporting to ONNX Runtime and TensorRT engine compilation for 5x GPU latency reduction
• Deploying streaming video inference servers via FastAPI and Triton Inference Server
• Docker containerization and deployment onto AWS SageMaker and NVIDIA Jetson edge boards

No. We provide dedicated cloud GPU sandboxes (NVIDIA T4 / A10G) and guide you on utilizing Google Colab Pro and Kaggle GPU runtimes for all assignments.

Familiarity with core Python programming and basic calculus/linear algebra concepts is recommended. We provide pre-course booster modules on PyTorch tensors and math fundamentals.

General ML focuses on tabular and statistical data. This track dives 100% into unstructured data (pixels, audio waveforms, video feeds, and natural language text) using modern deep foundation networks.