InvertibleNetworks.jl

InvertibleNetworks v2.2.9

Diff since v2.2.8

Merged pull requests:

• Minor changes in paper.md and fix Julia ref (#105) (@aurorarossi)
• Add minor changes to paper.md (#108) (@Nando-Hegemann)
• add memory benchmarks (#109) (@rafaelorozco)
• Cosmetic change readme (#110) (@rafaelorozco)
• Make minor readability suggestions (#111) (@drvinceknight)

Closed issues:

• Conditionally load heavy dependencies (#35)
• Wavelet squeeze is not GPU friendly (#52)
• Adding support + examples for FCN based INNs (#76)
• Improve visibility of contribution guidelines (#106)

InvertibleNetworks v2.2.8

Diff since v2.2.7

Merged pull requests:

• change vers and add cudnn dep (#107) (@rafaelorozco)

Closed issues:

• Error running conditional glow example on GPUs (#102)

InvertibleNetworks v2.2.7

Diff since v2.2.6

Merged pull requests:

• Get tests passing in Julia 10 (#103) (@rafaelorozco)

Closed issues:

• HINT layer input must be power of 2 in size? (#46)

InvertibleNetworks v2.2.6

Diff since v2.2.5

Merged pull requests:

• Dense glow network (#77) (@rafaelorozco)
• Summ net (#82) (@rafaelorozco)
• add posterior inference gaussian mu std example (#84) (@rafaelorozco)
• CompatHelper: bump compat for Wavelets to 0.10, (keep existing compat) (#90) (@github-actions[bot])
• add logdet option to glow and example of training with flux (#92) (@rafaelorozco)
• small spelling, add examples (#96) (@rafaelorozco)
• changes to docs to avoid runner crash (#97) (@rafaelorozco)
• change doc fig path for display (#98) (@rafaelorozco)
• add manual path to imgs (#99) (@rafaelorozco)
• bump version (#100) (@rafaelorozco)
• bump version correct (#101) (@rafaelorozco)

Closed issues:

• NetworkGlow has no field logdet (#80)
• Optional logdet in NetworkGlow (#81)

Summarizednetwork + docs
-summarized network structure beautifies the summary network workflow
-2 examples to docs will serve as template for many examples to come soon

InvertibleNetworks v2.2.5

Diff since v2.2.4

Closed issues:

• Flux friendlyness (#78)

Merged pull requests:

• conditional glow 3d w test and example (#66) (@rafaelorozco)
• fix Flux compat (#79) (@mloubout)

InvertibleNetworks v2.2.4

Diff since v2.2.3

Closed issues:

• Issues with forward evaluation of conv1x1 on GPU with CUDA 4 (#73)
• rrule error when ActNorm is nested with other network (#74)

Merged pull requests:

• Mix type rr (#75) (@mloubout)

InvertibleNetworks v2.2.3

Diff since v2.2.2

Closed issues:

• ridiculously long training times for GLOW on gpu (#72)

Merged pull requests:

• CompatHelper: bump compat for CUDA to 4, (keep existing compat) (#71) (@github-actions[bot])

InvertibleNetworks v2.2.2

Diff since v2.2.1

InvertibleNetworks v2.2.1

Diff since v2.2.0

Merged pull requests:

• Add the latest paper to README.md. (#63) (@alisiahkoohi)
• Update runtests.yml (#67) (@rafaelorozco)
• add activation parameter to conditional multiscale hint (#68) (@rafaelorozco)

InvertibleNetworks v2.2.0

Diff since v2.1.5

Merged pull requests:

• Conditional glow network and flexible residual block (#59) (@rafaelorozco)

InvertibleNetworks v2.1.5

Diff since v2.1.4

Merged pull requests:

• Glow 3d (#61) (@rafaelorozco)

InvertibleNetworks v2.1.4

Diff since v2.1.2

InvertibleNetworks v2.1.3

Diff since v2.1.2

InvertibleNetworks v2.1.2

Diff since v2.1.1

Closed issues:

• Reverse net has no rrule (#60)

Merged pull requests:

• Prevent unrealistic eager memory alloc in conv1x1 (my bad) (#53) (@mloubout)
• CompatHelper: bump compat for NNlib to 0.8, (keep existing compat) (#58) (@github-actions[bot])
• Dispatch pullback rrule for all Invertible (#62) (@ziyiyin97)

InvertibleNetworks v2.1.1

Diff since v2.1.0

Closed issues:

• glow example does not work (#47)
• Gradient calculation with Chainrule (#48)
• rename log_likelihood and Co.? (#50)

Merged pull requests:

• Fix dimension check in squeeze, (#43) (@rafaelorozco)
• Make sure example run and with good results (#49) (@rafaelorozco)
• log_likelihood: improve docstring (#51) (@alisiahkoohi)
• CompatHelper: bump compat for JOLI to 0.8, (keep existing compat) (#56) (@github-actions[bot])

InvertibleNetworks v2.1.0

Diff since v2.0.4

Merged pull requests:

• add julia 1.7 testing (#42) (@mloubout)

InvertibleNetworks v2.0.4

Diff since v2.0.3

Merged pull requests:

• make NetworkMultiScaleConditionalHINT reversible and use logdet correctly (#24) (@rafaelorozco)

InvertibleNetworks v2.0.3

Diff since v2.0.2

Merged pull requests:

• split_scales small bug (#39) (@rafaelorozco)
• add split_scales option to glow (#40) (@rafaelorozco)

InvertibleNetworks v2.0.2

Diff since v2.0.1

Closed issues:

• Release? (#32)
• Release type constraints (#34)

Merged pull requests:

• Fix signature types (#37) (@mloubout)
• add scale and shifted sigmoid (#38) (@rafaelorozco)