warming up your workspace

Python practice in Deep Learning

Browse the rooms before signing in. Opening a room requires an account and follows your existing access. Practice does not issue certificates.

  • ReLU Activation Audit

    ML engineer inspecting which supplied activations are zero and which are positive.

    networks. Free room.

  • Loss Alert

    On-call ML engineer: the training monitor reports 'loss = inf'.

    foundations. Free room.

  • Softmax Meltdown

    On-call ML engineer: the recommendation model is returning NaN probabilities.

    foundations. Free room.

  • Weight Initialization Review

    ML engineer replacing an all-zero initializer before training a classifier.

    networks. Free room.

  • Batch Stats Bug

    On-call ML engineer: image tagger gives inconsistent scores across a minibatch.

    training. Account access required.

  • Dense-Layer Gradient Review

    ML engineer checking a linear layer backward pass against its input and parameter shapes.

    backprop. Account access required.

  • Momentum State Review

    ML engineer checking a momentum update while preserving supplied optimizer state.

    training. Account access required.

  • Sigmoid Gradient Review

    ML engineer computing finite sigmoid derivatives for large positive and negative scores.

    backprop. Account access required.

  • Conv Crash

    On-call ML engineer: edge-detection step returns wrong-sized tensors.

    convnets. Account access required.

  • Embedding Lookup and Gradient Review

    ML engineer checking lookup copies and repeated-token gradient accumulation.

    embeddings. Account access required.

  • Precision, Recall and Threshold Review

    ML engineer comparing metrics and an explicitly defined lowest-feasible-threshold rule.

    classification. Account access required.

  • Recurrent State Review

    ML engineer tracing a tanh recurrent state through an input sequence.

    rnns. Account access required.

  • Attention Scaling and Mask Review

    ML engineer repairing attention values and masked-row behavior; sequence-memory optimization is outside this exercise.

    attention. Account access required.

  • CNN Backprop Bug

    On-call ML engineer: custom conv layer has broken gradient, blocking release.

    convnets. Account access required.

  • Gradient Clip Failure

    On-call ML engineer: language model fine-tune exploding on batch 3.

    training. Account access required.

  • Positional Encoding Regression

    On-call ML engineer: transformer outputs garbage for positions beyond 128.

    attention. Account access required.

  • BPTT Overflow

    On-call ML engineer: time-series classifier producing wrong BPTT gradients.

    rnns. Account access required.

  • Tied-Embedding Gradient Review

    ML engineer accumulating input-lookup and output-projection contributions to one shared embedding table.

    embeddings. Account access required.

  • Transformer Debugging: Forward, Loss and Context

    ML engineer debugging a small unmasked transformer, a stable loss and a cropped-context greedy decoder across three exercises.

    attention, rnns, foundations. Account access required.

  • Attention Temperature Review

    ML engineer inspecting how attention temperature changes weights and weighted values, without assuming improved answer calibration.

    attention. Account access required.