diff options
| author | haoyuren <13851610112@163.com> | 2026-07-28 07:53:50 +0800 |
|---|---|---|
| committer | haoyuren <13851610112@163.com> | 2026-07-28 07:53:50 +0800 |
| commit | 99b95217a30f75e9e35b5044850ced4ab802bc87 (patch) | |
| tree | 9aca82fc9f9165b6e7585eefe1061956638462aa /outputs/cnn_initialization_validation/cnn_initialization_rows.csv | |
| parent | e65dd2e2460b1da48c83a930304cf9b269fc4447 (diff) | |
Redesign main figure panels and track figure-source CSVs
Panel 1a: rename to Backward rules; DFA drawn as direct side-to-side
arrows from the error to each hidden layer. Panel 1b: replace the dual
y-axis with stacked guarantee/cost sub-axes. Figure 2: reader-facing
panel titles; legend renames (linear extrapolation, relinearization).
Track the regenerated figure-source CSVs (verified bit-exact against
the frozen paper numbers) so figures can be rebuilt without rerunning
experiments.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Diffstat (limited to 'outputs/cnn_initialization_validation/cnn_initialization_rows.csv')
| -rw-r--r-- | outputs/cnn_initialization_validation/cnn_initialization_rows.csv | 9 |
1 files changed, 9 insertions, 0 deletions
diff --git a/outputs/cnn_initialization_validation/cnn_initialization_rows.csv b/outputs/cnn_initialization_validation/cnn_initialization_rows.csv new file mode 100644 index 0000000..377aa5e --- /dev/null +++ b/outputs/cnn_initialization_validation/cnn_initialization_rows.csv @@ -0,0 +1,9 @@ +architecture,loss,rule,init_seed,feedback_draws,bp_speed,output_speed,predicted_deficit,empirical_deficit,empirical_stderr,empirical_std,calibration_error
+"Conv(8,12)-GAP-linear",cross_entropy,FA,10000,2048,0.13347306443714196,0.04703411830563681,0.6476134079637683,0.6481860017082137,0.005014764187885875,0.22694232085155533,0.0005725937444454043
+"Conv(8,12)-GAP-linear",cross_entropy,DFA,10000,2048,0.13347306443714196,0.04703411830563681,0.6476134079637683,0.6525109759254111,0.005114013686309232,0.2314338404268846,0.004897567961642757
+"Conv(8,12)-GAP-linear",cross_entropy,FA,10001,2048,0.15285396537155196,0.062427512417831094,0.5915872233599925,0.5998958349629218,0.00635323584180459,0.28751463335791677,0.008308611602929283
+"Conv(8,12)-GAP-linear",cross_entropy,DFA,10001,2048,0.15285396537155196,0.062427512417831094,0.5915872233599925,0.5877690993242832,0.0065504075513939134,0.2964376063440771,-0.003818124035709247
+"Conv(8,12)-GAP-linear",cross_entropy,FA,10002,2048,0.1010038649291602,0.02304187527553643,0.771871350747845,0.7775432310523354,0.0054593227304628724,0.24706074389735397,0.005671880304490395
+"Conv(8,12)-GAP-linear",cross_entropy,DFA,10002,2048,0.1010038649291602,0.02304187527553643,0.771871350747845,0.7762944617111598,0.00534470168890309,0.2418735876891244,0.004423110963314825
+"Conv(8,12)-GAP-linear",cross_entropy,FA,10003,2048,0.5008232453126622,0.09233667795914846,0.8156302072171131,0.8142268866524287,0.0051730785275724295,0.2341068100132471,-0.001403320564684396
+"Conv(8,12)-GAP-linear",cross_entropy,DFA,10003,2048,0.5008232453126622,0.09233667795914846,0.8156302072171131,0.8141641740382936,0.005155033671570825,0.23329019305041385,-0.001466033178819548
|