randomartist
You’re looking for collabortors?
I'm doing physics-inspired anomaly detetion & lookinf for a co-founder.
Willing to relocate: Yes, worldwide
Technologies: Python, Matlab Email: marthaelias [at] protonmail [dot] com Github: https://github.com/marthafay My Contribution (Short Profile):
- Applied Al research with focus on real-time signal processing, decision logic & anomaly detection - Modular signal & system architecture — complements classical ML stacks - SDKs: Audio, Finance - Physically informed analogy modules (e.g., superposition in Python) • Hand-crafted, explainable features & operators • Guiding principle: ML/RL cleanly embedded into a predefined architecture Example Result Phase-aware XGBoost trained on "Melodic House" harmonies → showed genre generality beyond training data Working Principle Reproducible research: clear pipelines, paper-style notebooks, deterministic exports Mainstream Deep Learning (TensorFlow) - Purpose: ML, Deep Learning, neural networks • Focus: Classification, regression, prediction • Architecture: Computation graph & tensors • Inputs: n-dim tensors • Outputs: Probabilities, models, vectors • Computational Principle: Gradient descent • Components: Layers, losses, optimizers • Tooling: GPU-first workflows, tf.keras, deployment stacks • Data Strategy: Big Data, automatic feature learning • Explainability: Often low (requires additional tooling) • Style: GPU-heavy, often overdimensioned • Versioning: Model checkpoints, weights
My Contribution (Short Profile) • Applied Al research with focus on real-time signal processing, decision logic & anomaly detection • Modular signal & system architecture — complements classical ML stacks • Physically informed analogy modules (e.g., superposition in Python) • Hand-crafted, explainable features & operators • Guiding principle: ML/RL cleanly embedded into a predefined architecture Example Result Phase-aware XGBoost trained on "Melodic House" harmonies → showed genre generality beyond training data Working Principle Reproducible research: clear pipelines, paper-style notebooks, deterministic exports Mainstream Deep Learning (TensorFlow) • Purpose: ML, Deep Learning, neural networks • Focus: Classification, regression, prediction • Architecture: Computation graph & tensors • Inputs: n-dim tensors • Outputs: Probabilities, models, vectors • Computational Principle: Gradient descent • Components: Layers, losses, optimizers • Tooling: GPU-first workflows, tf.keras, deployment stacks • Data Strategy: Big Data, automatic feature learning • Explainability: Often low (requires additional tooling) • Style: GPU-heavy, often overdimensioned • Versioning: Model checkpoints, weights