A researcher/engineer exploring the theoretical, mathematical, and infrastructural foundations of modern AI and Machine Learning by building them from scratch (pure NumPy, PyTorch & JAX). No high-level libraries; just pure math, algorithmic depth, and high-performance computing.
- mechanistic_interpretability: Reverse-engineering neural network internals from scratch (NumPy + PyTorch). Linear representation hypothesis, superposition, SAEs, induction heads, and causal scrubbing.
- deep_learning / derin_ogrenme: Everything about deep learning architecture and foundations.
- geometric_deep_learning: Manifolds, equivariance, symmetry groups, GNNs, and Riemannian geometry.
- graph_neural_networks: Spectral Graph Theory, graph Laplacian, over-smoothing, and raw GCN/GAT/GIN implementations.
-
calculus_for_ai / yapay_zeka_icin_kalkulus:
$\varepsilon–\delta$ limits, complex-step differentiation, and forward/reverse-mode autodiff from scratch in pure NumPy. - linear_algebra_for_ml / makine_ogrenmesi_icin_lineer_cebir: Geometric intuition, LU/QR decomposition, stable SVD, and PCA from scratch.
- optimization_methods: KKT duality, L-BFGS, proximal methods (ADMM), and cutting-edge deep learning optimizers (K-FAC, Shampoo, Muon).
- numerical_methods_for_ml / numerical_methods / sayisal_yontemler: Machine epsilon, Chebyshev interpolation, adaptive Gaussian quadrature, and stiff ODE simulators.
- discrete_mathematics / ayrik_matematik: Propositional logic, combinatorics, graph theory, and computational verification of theorems.
- probability & statistics / olasilik & istatistik: Theoretical background, Python implementations, and visualizations for engineers.
- differential_equations_for_ai / differential_equations / diferansiyel_denklemler: Neural ODEs, PINNs, diffusion models, and optimal control in NumPy + JAX.
- distributed_systems_for_ml: Custom Ring All-Reduce primitives, Parameter Servers, 1F1B Pipeline Parallelism, and ZeRO/FSDP memory sharding from scratch.
- high_performance_computing / hpc_ai_infra_llmops: Systems engineering required to train, fine-tune, and serve frontier-scale LLMs.
- mlops_and_deployment: Custom experiment trackers, containerized model registries, dynamic batching inference servers, and high-performance LLM serving infra.
- tensorlens: Visualizing Modern LLM Mechanics, Loss Landscapes & HPC Topologies.
- advanced_llm_architecture: Next-gen LLM components: Quantization, TTT layers, Differentiable Logic, and FlashAttention tiling.
- autonomous_ai_agents: Enterprise-grade AI Agents, MCP architecture, hierarchical indexing, and ReAct loops from scratch.
- generative_artificial_intelligence / uretken_yapay_zeka: Bridging statistical theory to deep generative architectures (VAE, GAN, Diffusion, and Mini-GPT with RoPE).
- nlp_course: Entire NLP stack from tokenization to PEFT/LoRA, DPO/RLHF alignment, and production RAG.
- causal_inference_for_ai: d-separation oracles, propensity score IRLS, doubly-robust AIPW, and Cross-Fitting Double ML (DML).
- bayesian_machine_learning: Laplace approximations, from-scratch MCMC (HMC/NUTS), Variational Inference (ELBO), and Sparse GPs.
-
reinforcement_learning / pekistirmeli_ogrenme: From MDP foundations to TD(
$\lambda$ ), Q-Learning, and Policy Gradients (REINFORCE, Actor-Critic). - time_series_analysis / zaman_serisi_analizi: Spectral Analysis, ARCH/GARCH, Multivariate VAR, and State-Space models with Kalman Filters.
- prod_grade_tab_ml: Custom GBDT boosters and inner mechanics of XGBoost, LightGBM, and CatBoost.
- data_mining / veri_madenciligi: Advanced classification, imbalanced data, Association Analysis, and Anomaly detection.
- quantum_artificial_intelligence: Hands-on quantum computing for machine learning, built from scratch in NumPy then bridged to PennyLane. Covers qubits & gates, entanglement & CHSH, quantum algorithms (Deutsch–Jozsa, Grover, QFT), variational circuits & VQE, quantum ML classifiers, and quantum kernels.
- computer_vision / bilgisayarli_goru: No OpenCV, no PyTorch. Custom convolutions, Canny edge detection, Harris corners, Lucas–Kanade optical flow, and custom CNN layers.
- adv_pde_based_image_processing: Advanced Geometric Image Processing, Finsler/Randers metric flows with Numba/CUDA hardware acceleration.
- Riemannian-Curve-Evolution: Riemannian Curve Model Analysis: Image Segmentation Application.
- A-Novel-Family-of-Edge-Preserving-Anisotropic-Filters: A Novel Family of Edge Preserving Anisotropic Filters.
- data_vis_for_ai_research: High-dimensional embeddings (UMAP), model diagnostics, and Explainable AI (SHAP, LIME, Grad-CAM).
- Languages: Python, C++, CUDA
- Math & Core ML: Pure NumPy, JAX, PyTorch, PennyLane, Numba
- Systems & Operations: Distributed Training, HPC Topologies, MLOps, Triton/LLM Serving Infra
- LinkedIn: https://linkedin.com/in/haydarkilicai
- Email: haydarkilicinfo@gmail.com