Machine Psychological Bureau

A M S T E R D A M

Machine Learning & Internal SEO (Research/Prototyping/Optimisation):

Supervised: Classification (e.g: Customer, Sentimental) & Regression Tasks (e.g Time Series forecasting, LTV)

Unsupervised: Dimensionality Reduction (Meaningful, Structure, Features) & Clustering (Recommender, Customer)

Reinforcement: Temporal Difference (TD), Monte Carlo (MC)

(Glove, PCA/t-SNE/UMAP, LTR (WALS), BigQuery-ML, TFHub)

Model Converter TF to TFJS

Motivation: Port a TF Actor-Critic (CartPole-v1) GraphModel (23.1 kB) to TFJS and run inference in the web . The web interface allows easily to manipulate the environment and sensor data input for realtime behavioural testing.
Glitch Browser Demo

PWA: Lighthouse optimised model loading

Predicting mousewritten digit's in realtime. The CNN model was trained in Python/TF, converted to TFJS and optimised for progressive loading via service worker. "Lighthouse Fireworks"

Try the Browser Demo

Testing different TFA's on Gym Classics

Rendered video results of benchmarking Gym Classics with TFA.

Enjoy the Webstory

Keyword Classifier

Train a little ML model to classify keywords (GSC, Adwords, Trends) into multiple categories

Find out more

Deep Learning (Research/Optimisation):

GD, (k)NN, CNN, RNN, DCN, GNN (GCN), RL, DQN, AC, SAC

(NumPy, Keras, Tensorflow (TFJS, TFRS, TFA, TFP, ...), Spektral, ConvNetJS)

3d_argmax_heatmap.mp4

3d Prediction Heat-map

The animation show's the confidence values of 8 million predictions in the given dataset.

3d_predictions.mp4

2 hidden layer softmax classifier

Coded a FCMLP in NumPy, the animation show's the final classification of a 3d spiral.

Feel free to have a look at Github.

Game-Crawler-MRP-AD.mp4

Solving MRP using Bellman's equation

The decision process was visualised by graphVis using the Bellman equation recursively for solving a little game MRP coded in Python (NumPy).

Recommender & Internal SEO (Research/Optimisation):

Internal Search Engines: Crawling, Indexing and Ranking. Searching by Query > Document/ID
Query-less Recommender: Crawling, Indexing and Ranking. Searching by [User, Image, Audio] > Document/ID
Multilingual Recommender: Crawling, Indexing [optional: Ranking]. Searching by [DE] Query > [15+ languages] Document/ID

(TFRS, TF, Pointwise-, Pairwise- and Listwise Rankings)

Multilingual Document Recommender

Motivation: Using convergences between TF, TFRS and TFhub. Creating a little Search Index Playground to find all multilingual Twins (semantic similar).

Find out more Use Cases

Page updated

Google Sites

Report abuse