Machine Learning Toolkit

Machine Learning Toolkit: Skills and Expertise #

Data Engineering #

  • Expertise in building unbiased datasets via feature-based sampling.
  • Proficient in generating synthetic data matching real data distribution.
  • Skilled in augmentation techniques for vision, speech, NLP.

Modelling and Feature Engineering #

  • Comprehensive knowledge of Convolutional, Recurrent, and Transformer-based models.
  • Experience with feature importance techniques.
  • Proficient with Contrastive Learning methods like SimCLR, BYOL, SimSiam.
  • Well-versed in model debugging and profiling.
  • Experienced with topic models using Probabilistic Graphic Models and embedding-based clustering.

Training #

  • Proficient in distributed training using OpenMPI + RoCE, Torch RPC.
  • Skilled in Pytorch Lightning optimizations.

Calibration #

  • Expertise in implicit calibration techniques like Focal Loss, Maximum Entropy Regularization, Label Smoothing, Random Dropout.
  • Experience with explicit calibration techniques like Isotonic Regression, Platt’s scaling.

Optimizations #

  • Skilled in model optimizations such as Quantization, Pruning, Distillation.
  • Proficient in ML Ops Fusing techniques like ONNX, TorchDynamo, TVM.

Inference #

  • Expertise in C++ inference using ONNX & Drogon.
  • Experience with frameworks like Triton, Mosec.
  • Skilled in scaling on k8 using OKD.
  • Proficient in monitoring and alerting using Vector.io, Prometheus, Grafana.

Online Monitoring #

  • Expertise in hard negative mining around calibrated threshold region.
  • Experience with sampling and saving hard negatives.
  • Skilled in detecting and alerting on Model and Data Drifts.