NIF
  • Neural Implicit Flow (NIF): mesh-agnostic dimensionality reduction
    • Highlights
      • Installation and requirements
      • Configuring and training NIF
      • Distributed learning
      • Fine-tuning with L-BFGS
      • Training schedules and monitoring
      • Extracting subnetworks
      • Input-output Jacobians and Hessians
      • Data normalization
      • Large-scale record data
      • Save and load models
      • Network pruning and quantization
    • Google Colab Tutorial
    • Issues, bugs, requests, ideas
    • How to cite
    • Contributors
    • License
  • Eight scientific tutorials
    • Original cases and compatibility notebooks
  • Distributed training and TensorBoard example
    • Run from a checkout
    • Training and output semantics
    • Verification scope
  • Models
    • Configuration and precision
    • Parameter-conditioned MLP field
      • NIF
    • Multiscale SIREN field
      • NIFMultiScale
    • Linear spatial representation
      • NIFMultiScaleLastLayerParameterized
  • Layers and derivatives
    • Pointwise derivatives
      • JacobianLayer
      • HessianLayer
      • compute_output_and_grad()
      • compute_output_and_grad_and_hessian()
      • compute_output_and_augment_grad()
    • Latent and activation penalties
      • JacRegLatentLayer
      • ParameterOutputL1ActReg
    • SIREN and hypernetwork layers
      • SIREN
      • SIREN_ResNet
      • HyperLinearForSIREN
      • compute_number_of_weightbias_by_its_position_for_shapenet()
      • gen_hypernetwork_weights_bias_for_siren_shapenet()
    • Dense and residual helpers
      • nif.layers.Dense
      • MLP_ResNet
      • MLP_SimpleShortCut
      • EinsumLayer
      • BiasAddLayer
  • Optimizers
    • First-order methods
      • ReferenceAdam
      • AdaBeliefOptimizer
      • Lion
      • L4Adam
    • L-BFGS refinement
      • ReferenceLBFGS
    • Packed objective helpers
      • lbfgs_minimize()
      • LBFGSOptimizer
      • TFPLBFGS
      • function_factory()
      • create_value_and_gradient_function()
      • pack_tensors()
      • unpack_tensors()
      • assign_tensors()
    • Gradient centralization
      • centralized_gradients_for_optimizer()
      • get_centralized_gradients()
  • Data and normalization
    • Point-wise data
      • PointWiseData
    • Streamed record datasets
      • TFRDataset
  • Packaged datasets
    • Traveling wave
      • TravelingWave
    • High-frequency traveling wave
      • TravelingWaveHighFreq
    • Cylinder flow
      • CylinderFlow
  • Magnitude pruning
    • PolynomialMagnitudePruning
      • PolynomialMagnitudePruning.mask
      • PolynomialMagnitudePruning.step
      • PolynomialMagnitudePruning.refresh_count
      • PolynomialMagnitudePruning.on_epoch_end()
      • PolynomialMagnitudePruning.strip()
    • Lifecycle
  • Native API and intentional migration corrections
    • Models and ordering
    • Explicit regularization
    • Derivatives
    • Reference Adam epsilon convention
    • Native optimizers and streaming data
    • Tutorials and artifacts
  • Native PyTorch validation
    • Independent package acceptance
    • Full tutorial execution
    • Setup and hosted Colab
  • Controlled TensorFlow/PyTorch tutorial comparison
    • Independence and source preservation
    • Controls and progression
    • Tutorial-specific checks
    • Current evidence
      • Tutorial 3: user acceptance
      • First complete controlled tutorial 3 pair
      • Candidate 016: complete tutorial 3 revalidation still fails
      • Tutorial 1: first paired initialization through both optimizers
      • Tutorial 1: three original own-stage pairs
      • Tutorial 1: reference L-BFGS contract repair, CPU acceptance
      • Tutorial 1: repaired optimizer completes the matrix, with failures retained
      • Tutorial 1: official PyTorch L-BFGS quality comparison
      • Tutorial 4 derivative deliverables
      • Tutorial 5: three complete controlled pairs and reference repeat
      • Tutorial 2: user acceptance
      • Tutorial 2: first complete pair fails the original field criterion
      • Tutorial 2: all three complete paired initializations
      • Tutorial 2: declared-state disk controls complete
      • Tutorial 2: descriptive variation across retained seeds
      • Tutorial 7: complete baselines fail before primary pruning
      • Tutorial 6: GPU controls retain cross-framework failures
      • Tutorial 8: controls and complete matched training
      • Learning tutorials: execution, numerical results and user acceptance
      • Hosted Colab tutorial 1: full execution, separate from parity
      • Hosted Colab tutorial 2: both complete fits
      • Hosted Colab tutorial 3: complete cylinder-flow workload
      • Hosted Colab tutorial 4: derivative notebook execution
      • Hosted Colab tutorial 5: complete streamed workload
      • Hosted Colab tutorial 6: all nine regularization fits
      • Hosted Colab tutorial 7: pruning and native int8 export
      • Hosted Colab tutorial 8: both complete Sobolev and baseline fits
      • Public regression portability
NIF
  • Search

© Copyright 2026, Shaowu Pan, Steven Brunton, J. Nathan Kutz.

Built with Sphinx using a theme provided by Read the Docs.