From c24a6a575bce15e7d7da541ecb752c65b5e68e47 Mon Sep 17 00:00:00 2001 From: YurenHao0426 Date: Wed, 22 Jul 2026 13:22:24 -0500 Subject: results: close fixed hierarchical FA short gate --- .gitignore | 1 + BASELINES.md | 3 +- HFA_BASELINE.md | 13 + RESULTS.md | 26 ++ REVIEW_SCORECARD.md | 2 + ROADMAP.md | 12 + results/oral_a_hfa_short/hfa_lr0.01.json | 534 +++++++++++++++++++++++++++++++ results/oral_a_hfa_short/hfa_lr0.03.json | 534 +++++++++++++++++++++++++++++++ results/oral_a_hfa_short/hfa_lr0.1.json | 534 +++++++++++++++++++++++++++++++ results/oral_a_hfa_short_selection.json | 69 ++++ 10 files changed, 1727 insertions(+), 1 deletion(-) create mode 100644 results/oral_a_hfa_short/hfa_lr0.01.json create mode 100644 results/oral_a_hfa_short/hfa_lr0.03.json create mode 100644 results/oral_a_hfa_short/hfa_lr0.1.json create mode 100644 results/oral_a_hfa_short_selection.json diff --git a/.gitignore b/.gitignore index d2fffea..92067a7 100644 --- a/.gitignore +++ b/.gitignore @@ -5,6 +5,7 @@ __pycache__/ # Runtime logs are reproducible and can grow quickly. logs/ *.out +*.log # Local editor/OS state. .DS_Store diff --git a/BASELINES.md b/BASELINES.md index 9800d16..6b76d09 100644 --- a/BASELINES.md +++ b/BASELINES.md @@ -20,7 +20,8 @@ The audited implementations and completed results are in `RESULTS.md`: adjoints, and local ReLU/BatchNorm Jacobians; it never copies or reads a downstream forward convolution and is the mandatory baseline for any learned hierarchical extension. Its frozen validation-only screen is specified in - `HFA_BASELINE.md`; + `HFA_BASELINE.md`; the selected short endpoint is 43.52% versus matched DFA's + 37.16%, but its 50% full-run gate failed; - direct node perturbation uses a causal target on every hidden update and no learned vectorizer; - the no-traffic, `P=0` SDIL backbone is learned direct feedback by node diff --git a/HFA_BASELINE.md b/HFA_BASELINE.md index 1dc5f87..64d9c22 100644 --- a/HFA_BASELINE.md +++ b/HFA_BASELINE.md @@ -62,3 +62,16 @@ test-set evaluation. - HFA-S2, if opened, copies the selected S1 hidden rate mechanically and does not select on intermediate validation accuracy. +## Audited outcome (2026-07-22) + +All three HFA-S1 trajectories are finite and test-free. Validation accuracy +increases monotonically over the frozen hidden-rate grid: `39.64%` at 0.01, +`41.58%` at 0.03, and `43.52%` at 0.1. The selected 0.1 run has early-third +teaching alignment `0.040406`, zero causal queries, 267,904 fixed feedback +parameters, and estimated work `2.4242e13` MACs. + +The selected endpoint exceeds matched DFA by `6.36` points and the failed-v1 +short SDIL endpoint by `1.54` points, but misses the preregistered 50% HFA-S2 +threshold by `6.48` points. Status is `selected_full_closed`; the 200-epoch HFA +run is not launched, no extra HFA setting is allowed, and CIFAR-10 test remains +untouched. diff --git a/RESULTS.md b/RESULTS.md index eb8feec..d13a58c 100644 --- a/RESULTS.md +++ b/RESULTS.md @@ -728,6 +728,32 @@ must learn these hierarchical feedback maps without copying forward weights and must compare against fixed/random hierarchical FA and BurstCCN; merely adding recursive feedback is established prior art. +### Matched convolutional hierarchical-FA baseline + +The HFA implementation was mechanically audited and its three-run short screen +was frozen before any HFA accuracy endpoint was observed. It mirrors the real +option-A residual DAG with independent random 3x3 feedback tensors and local +ReLU/BatchNorm Jacobians. An audit-only symmetric copy reproduces exact hidden +credit to `4.97e-16` relative error and the BP parameter update to `1.49e-8` +absolute error; actual runs never copy a forward tensor. + +On the identical 10k-example, 20-epoch ResNet-20 A2b setting: + +| hidden LR | validation accuracy | early-third alignment | estimated MACs | causal queries | +|--:|--:|--:|--:|--:| +| 0.01 | 39.64% | 0.004572 | 2.4242e13 | 0 | +| 0.03 | 41.58% | 0.011338 | 2.4242e13 | 0 | +| 0.10 | **43.52%** | **0.040406** | 2.4242e13 | 0 | + +All three trajectories are finite, use source commit `f37644e`, allocate +1.731 GB peak GPU memory, take 89.3--89.6 seconds on one GTX 1080, and never +evaluate test. Selected HFA improves on matched tuned DFA (`37.16%`) by 6.36 +points and failed-v1 SDIL (`41.98%`) by 1.54 points, but remains 31.42 points +below BP (`74.94%`). It misses the frozen 50% full-run threshold by 6.48 +points, so HFA-S2 is closed without further tuning. The result confirms that +spatial hierarchy is useful but that fixed random hierarchy alone does not +solve standard-depth credit. + ## How to run `experiments/run.py --mode {bp,fa,dfa,sdil} --dataset {mnist,fmnist,cifar10} --depth D --residual {0,1} --act {tanh,gelu,silu,relu}` Batteries: `experiments/run_v2.sh "" "" `. diff --git a/REVIEW_SCORECARD.md b/REVIEW_SCORECARD.md index 60c58d3..bbdb604 100644 --- a/REVIEW_SCORECARD.md +++ b/REVIEW_SCORECARD.md @@ -60,6 +60,7 @@ Every formal result report records: | Oral-A A3 fails | 5 | Full ResNet-20 SDIL became nonfinite at epoch 89 and ended at 10%; DFA ended finite at 33.06% | Standard-scale and oral-A claims are closed; A4 remains untouched | | Oral-A-v2 causal capture fails | 5 | Structured perturbation improves early/all-layer alignment 6.5x/4.5x but misses the frozen early gate | Mechanism diagnosis sharpens; no full standard-scale endpoint opens | | Oral-A-v3 causal capture fails | 5 | Direct A/G perturbation lowers synthetic estimator variance but leaves real early alignment at 0.0071 | Coefficient regression is not the dominant early bottleneck; full scale remains closed | +| Fixed hierarchical FA short gate fails | 5 | HFA reaches 43.52%, above DFA 37.16% and failed-v1 SDIL 41.98%, but below its frozen 50% full-run threshold | Spatial hierarchy helps; fixed random hierarchy remains insufficient and supplies no SDIL scale evidence | | Oral-A A4 | not opened | The prerequisite A3 gate failed | No oral-A confirmation claim is available | These are conditional reviewer forecasts, not promised scores. A failed stage leaves its negative @@ -78,6 +79,7 @@ retroactively reopened by success on standard vision benchmarks. | 2026-07-22 / post-failure representation oracle | Current family has a 0.0240 cross-validated early-alignment ceiling on the fixed probe versus 0.0072 learned; unconstrained basis coefficients reach 0.0549 | 5 → 5 | Localizes both a causal-regression gap and an output-error-only capacity gap, but an oracle audit supplies no task-performance evidence | | 2026-07-22 / Oral-A-v3 V3-1 | Four clean frozen-forward records: vectorizer-space estimation raises all-layer alignment to 0.0626 but leaves early alignment at 0.0071 and fails three advancement checks | 5 → 5 | Exact lower-variance mechanics do not repair early credit; no full ResNet or confirmation evidence is opened | | 2026-07-22 / hierarchical oracle | Held-out gated 3x3 maps from exact child error fields reach 0.9998 early alignment; ordinary activation context remains 0.0321 | 5 → 5 | Strongly localizes the missing information to spatial hierarchical error fields, but exact-gradient oracle inputs provide no evidence that the proposed local learner can obtain them | +| 2026-07-22 / fixed HFA S1 | Three clean matched ResNet-20 records; selected HFA reaches 43.52% and 0.0404 early alignment but misses the frozen 50% gate | 5 → 5 | Establishes a stronger zero-query local baseline and confirms hierarchy helps, while closing an uncalibrated full run and leaving standard SDIL evidence absent | Future rows are appended only after an audited frozen stage. A score staying flat is informative: engineering, theory exposition, or visualization may make the paper more defensible without diff --git a/ROADMAP.md b/ROADMAP.md index 15699e2..d6bfa83 100644 --- a/ROADMAP.md +++ b/ROADMAP.md @@ -392,6 +392,18 @@ and uses exact child gradients as oracle inputs. It establishes that future development should learn spatial hierarchical feedback without weight copying, with random/fixed hierarchical FA and BurstCCN as mandatory baselines. +**Fixed hierarchical-FA baseline status: short gate failed.** The audited +ResNet-20 screen reaches `39.64%`, `41.58%`, and `43.52%` at hidden rates 0.01, +0.03, and 0.1. The selected run improves on matched DFA by 6.36 points and has +early alignment `0.040406`, but misses its frozen 50% full-run threshold. +HFA-S2 is therefore closed. This validates hierarchical feedback as a stronger +baseline while ruling out fixed random hierarchy as the complete repair. + +The next mechanics branch perturbs every hierarchical feedback-parameter +subspace in one antithetic pair and subtracts locally computed prediction +moments. Its causal JVP and exact symmetric-limit delta rule pass, but it has no +task claim until a frozen-forward alignment gate is preregistered and audited. + Prepare convolutional local-update primitives and ResNet-20/32/56 protocols early. Queue frozen runs opportunistically on authorized idle GPUs. Because BurstCCN already reports CIFAR-10 and ImageNet scaling, dataset scale alone is not novel. The oral-level target is a memorable joint diff --git a/results/oral_a_hfa_short/hfa_lr0.01.json b/results/oral_a_hfa_short/hfa_lr0.01.json new file mode 100644 index 0000000..32a1365 --- /dev/null +++ b/results/oral_a_hfa_short/hfa_lr0.01.json @@ -0,0 +1,534 @@ +{ + "architecture": { + "adaptive_apical_parameters": 0, + "base_width": 16, + "blocks_per_stage": 3, + "bn_eps": 1e-05, + "bn_momentum": 0.1, + "depth": 20, + "family": "CIFAR 6n+2 ResNet, option-A shortcuts", + "fixed_feedback_parameters": 267904, + "fixed_traffic_coefficients": 0, + "forward_parameters": 269722, + "hidden_shapes": [ + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ] + ], + "normalization": "batchnorm", + "predictor_parameters": 0, + "residual_scale": 1.0, + "vectorizer_mode": null, + "vectorizer_parameters": 0 + }, + "args": { + "a_scale": 1.0, + "a_warmup_steps": 0, + "alignment_probe": 32, + "apical_calibration_mode": "unit_targets", + "apical_seed": null, + "augment_train": 1, + "batch_size": 128, + "bn_eps": 1e-05, + "bn_momentum": 0.1, + "data_dir": "/home/yurenh2/sdrn/data", + "depth": 20, + "device": "cuda", + "epochs": 20, + "eta_A": 0.01, + "eta_P": 0.01, + "eval_every": 0, + "eval_split": "validation", + "learn_P": 0, + "loader_seed": 0, + "lr": 0.01, + "lr_gamma": 0.1, + "lr_milestones": "100,150", + "lr_schedule": "cosine", + "max_steps": 0, + "mode": "hfa", + "momentum": 0.9, + "normalization": "batchnorm", + "nuisance_scale": 0.0, + "out": "results/oral_a_hfa_short/hfa_lr0.01.json", + "output_lr": 0.1, + "pert_directions": 1, + "pert_every": 4, + "pert_sigma": 0.01, + "perturb_seed": 1000, + "predictor_warmup_steps": 0, + "residual_scale": null, + "seed": 0, + "split_seed": 2027, + "train_limit": 10000, + "use_residual": 1, + "val_examples": 5000, + "vectorizer_mode": "spatial_template", + "warmup_epochs": 0, + "weight_decay": 0.0001, + "weight_scale": 1.0, + "width": 16 + }, + "calibration_metric_space": null, + "counters": { + "apical_warmup_examples": 0, + "calibration_event_examples": 0, + "causal_scalar_observations": 0, + "logical_batch_loss_queries": 0, + "ordinary_examples": 200000, + "per_example_loss_terms": 0, + "perturbation_events": 0, + "perturbation_forward_examples": 0, + "predictor_warmup_examples": 0 + }, + "diagnostics": { + "early_third_mean": 0.004572224900281678, + "innovation_negative_gradient_cosine": [ + 0.0020579462870955467, + 0.0012284447439014912, + 0.007407092023640871, + 0.00022282346617430449, + 0.016841813921928406, + -0.0003247710410505533, + 0.030386799946427345, + 0.0017005586996674538, + 0.060705218464136124, + 0.015753068029880524, + 0.09917206317186356, + 0.0011747075477614999, + 0.15302309393882751, + 0.006798408459872007, + 0.21235091984272003, + 0.10412535071372986, + 0.2994855046272278, + 0.13240748643875122, + 0.3866526484489441 + ], + "normalization_state": "training_batch_stats_without_running_update", + "raw_negative_gradient_cosine": [ + 0.0020579462870955467, + 0.0012284447439014912, + 0.007407092023640871, + 0.00022282346617430449, + 0.016841813921928406, + -0.0003247710410505533, + 0.030386799946427345, + 0.0017005586996674538, + 0.060705218464136124, + 0.015753068029880524, + 0.09917206317186356, + 0.0011747075477614999, + 0.15302309393882751, + 0.006798408459872007, + 0.21235091984272003, + 0.10412535071372986, + 0.2994855046272278, + 0.13240748643875122, + 0.3866526484489441 + ], + "teaching_negative_gradient_cosine": [ + 0.0020579462870955467, + 0.0012284447439014912, + 0.007407092023640871, + 0.00022282346617430449, + 0.016841813921928406, + -0.0003247710410505533, + 0.030386799946427345, + 0.0017005586996674538, + 0.060705218464136124, + 0.015753068029880524, + 0.09917206317186356, + 0.0011747075477614999, + 0.15302309393882751, + 0.006798408459872007, + 0.21235091984272003, + 0.10412535071372986, + 0.2994855046272278, + 0.13240748643875122, + 0.3866526484489441 + ], + "wall_s": 0.12525105476379395 + }, + "epochs": [ + { + "epoch": 1, + "lr": 0.01, + "output_lr": 0.1, + "step": 79, + "train_examples": 10000, + "train_loss": 2.7393385593414306 + }, + { + "epoch": 2, + "lr": 0.009938441702975689, + "output_lr": 0.0993844170297569, + "step": 158, + "train_examples": 10000, + "train_loss": 2.3863267654418947 + }, + { + "epoch": 3, + "lr": 0.009755282581475769, + "output_lr": 0.09755282581475769, + "step": 237, + "train_examples": 10000, + "train_loss": 2.37835657081604 + }, + { + "epoch": 4, + "lr": 0.00945503262094184, + "output_lr": 0.09455032620941839, + "step": 316, + "train_examples": 10000, + "train_loss": 2.443051778793335 + }, + { + "epoch": 5, + "lr": 0.009045084971874737, + "output_lr": 0.09045084971874738, + "step": 395, + "train_examples": 10000, + "train_loss": 2.3301113273620606 + }, + { + "epoch": 6, + "lr": 0.008535533905932738, + "output_lr": 0.08535533905932738, + "step": 474, + "train_examples": 10000, + "train_loss": 2.2099775257110594 + }, + { + "epoch": 7, + "lr": 0.007938926261462366, + "output_lr": 0.07938926261462366, + "step": 553, + "train_examples": 10000, + "train_loss": 2.256035614776611 + }, + { + "epoch": 8, + "lr": 0.007269952498697734, + "output_lr": 0.07269952498697733, + "step": 632, + "train_examples": 10000, + "train_loss": 2.191825396156311 + }, + { + "epoch": 9, + "lr": 0.006545084971874737, + "output_lr": 0.06545084971874737, + "step": 711, + "train_examples": 10000, + "train_loss": 2.0189443355560304 + }, + { + "epoch": 10, + "lr": 0.0057821723252011546, + "output_lr": 0.05782172325201155, + "step": 790, + "train_examples": 10000, + "train_loss": 1.9209733907699584 + }, + { + "epoch": 11, + "lr": 0.005, + "output_lr": 0.05, + "step": 869, + "train_examples": 10000, + "train_loss": 1.9395836170196534 + }, + { + "epoch": 12, + "lr": 0.004217827674798845, + "output_lr": 0.04217827674798845, + "step": 948, + "train_examples": 10000, + "train_loss": 1.8563231918334961 + }, + { + "epoch": 13, + "lr": 0.003454915028125263, + "output_lr": 0.03454915028125263, + "step": 1027, + "train_examples": 10000, + "train_loss": 1.8415284217834473 + }, + { + "epoch": 14, + "lr": 0.0027300475013022664, + "output_lr": 0.027300475013022664, + "step": 1106, + "train_examples": 10000, + "train_loss": 1.816683451461792 + }, + { + "epoch": 15, + "lr": 0.0020610737385376348, + "output_lr": 0.02061073738537635, + "step": 1185, + "train_examples": 10000, + "train_loss": 1.7413049175262452 + }, + { + "epoch": 16, + "lr": 0.0014644660940672626, + "output_lr": 0.014644660940672627, + "step": 1264, + "train_examples": 10000, + "train_loss": 1.7405946004867554 + }, + { + "epoch": 17, + "lr": 0.0009549150281252633, + "output_lr": 0.009549150281252633, + "step": 1343, + "train_examples": 10000, + "train_loss": 1.72834784450531 + }, + { + "epoch": 18, + "lr": 0.0005449673790581611, + "output_lr": 0.005449673790581611, + "step": 1422, + "train_examples": 10000, + "train_loss": 1.7004222576141357 + }, + { + "epoch": 19, + "lr": 0.00024471741852423234, + "output_lr": 0.0024471741852423235, + "step": 1501, + "train_examples": 10000, + "train_loss": 1.686814351272583 + }, + { + "epoch": 20, + "lr": 6.15582970243117e-05, + "output_lr": 0.0006155829702431171, + "step": 1580, + "train_examples": 10000, + "train_loss": 1.6884625955581665 + } + ], + "evaluation_protocol": { + "test_evaluations": 0, + "test_used_for_selection": false, + "validation_evaluations": 1 + }, + "final": { + "accuracy": 0.3964, + "epoch": 20, + "evaluation_split": "validation", + "finite": true, + "loss": 1.6798031494140624, + "step": 1580 + }, + "hardware": { + "cuda_device_name": "NVIDIA GeForce GTX 1080", + "cuda_visible_devices": "5", + "device": "cuda", + "device_total_memory_bytes": 8507949056, + "peak_memory_allocated_bytes": 1730831872, + "peak_memory_reserved_bytes": 2390753280, + "torch_version": "2.3.1+cu118" + }, + "protocol_family": "oral_a_cifar_local_resnet_development", + "provenance": { + "git_commit": "f37644eb452f0eb3d7f368740f8b39493f11b2aa", + "git_tracked_dirty": false + }, + "schema_version": 1, + "split": { + "cifar_source_files": [ + { + "bytes": 31035704, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_1", + "sha256": "54636561a3ce25bd3e19253c6b0d8538147b0ae398331ac4a2d86c6d987368cd" + }, + { + "bytes": 31035320, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_2", + "sha256": "766b2cef9fbc745cf056b3152224f7cf77163b330ea9a15f9392beb8b89bc5a8" + }, + { + "bytes": 31035999, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_3", + "sha256": "0f00d98ebfb30b3ec0ad19f9756dc2630b89003e10525f5e148445e82aa6a1f9" + }, + { + "bytes": 31035696, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_4", + "sha256": "3f7bb240661948b8f4d53e36ec720d8306f5668bd0071dcb4e6c947f78e9682b" + }, + { + "bytes": 31035623, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_5", + "sha256": "d91802434d8376bbaeeadf58a737e3a1b12ac839077e931237e0dcd43adcb154" + }, + { + "bytes": 31035526, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/test_batch", + "sha256": "f53d8d457504f7cff4ea9e021afcf0e0ad8e24a91f3fc42091b8adef61157831" + } + ], + "dataset": "cifar10", + "evaluation_split": "validation", + "input_layout": "NCHW", + "input_shape": [ + 3, + 32, + 32 + ], + "loader_seed": 0, + "normalization_mean": [ + 0.49140000343322754, + 0.4821999967098236, + 0.4465000033378601 + ], + "normalization_std": [ + 0.24699999392032623, + 0.2434999942779541, + 0.26159998774528503 + ], + "split_from_training_only": true, + "split_seed": 2027, + "test_examples": 10000, + "train_examples": 10000, + "training_augmentation": "random_crop_32_padding_4_zero_then_horizontal_flip_p0.5", + "validation_class_counts": { + "0": 500, + "1": 500, + "2": 500, + "3": 500, + "4": 500, + "5": 500, + "6": 500, + "7": 500, + "8": 500, + "9": 500 + }, + "validation_examples": 5000, + "validation_index_sha256": "8328b206a97c420e49e54e3eca4abe3274c4756b084355784ea3fb8059e4515b" + }, + "timing": { + "apical_warmup_wall_s": 0.0, + "evaluation_wall_s": 0.36951661109924316, + "predictor_warmup_wall_s": 0.0, + "timing_excludes_data_loading_hashing_and_model_construction": true, + "total_timed_wall_s": 89.53109097480774, + "train_wall_s": 89.15893292427063 + }, + "work": { + "apical_macs_per_example": 40108672, + "causal_scalar_observations": 0, + "components": { + "apical_projection_macs": 8021734400000, + "apical_regression_macs": 0, + "bp_reverse_macs_estimate": 0, + "local_weight_correlation_macs": 8110208000000, + "ordinary_forward_macs": 8110208000000, + "perturbation_forward_macs": 0, + "warmup_clean_forward_macs": 0 + }, + "definition": "multiply-accumulates in conv/linear maps; one local weight correlation equals one forward-weight MAC count; BP reverse is estimated as one weight-gradient plus one activation-gradient convolution per forward convolution; elementwise nonlinearities and optimizer arithmetic excluded", + "forward_macs_per_example": 40551040, + "logical_batch_loss_queries": 0, + "per_example_cross_entropy_terms": 0, + "total_clean_forward_examples": 200000, + "total_forward_equivalent_examples": 200000, + "total_macs_estimate": 24242150400000 + } +} diff --git a/results/oral_a_hfa_short/hfa_lr0.03.json b/results/oral_a_hfa_short/hfa_lr0.03.json new file mode 100644 index 0000000..113ea64 --- /dev/null +++ b/results/oral_a_hfa_short/hfa_lr0.03.json @@ -0,0 +1,534 @@ +{ + "architecture": { + "adaptive_apical_parameters": 0, + "base_width": 16, + "blocks_per_stage": 3, + "bn_eps": 1e-05, + "bn_momentum": 0.1, + "depth": 20, + "family": "CIFAR 6n+2 ResNet, option-A shortcuts", + "fixed_feedback_parameters": 267904, + "fixed_traffic_coefficients": 0, + "forward_parameters": 269722, + "hidden_shapes": [ + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ] + ], + "normalization": "batchnorm", + "predictor_parameters": 0, + "residual_scale": 1.0, + "vectorizer_mode": null, + "vectorizer_parameters": 0 + }, + "args": { + "a_scale": 1.0, + "a_warmup_steps": 0, + "alignment_probe": 32, + "apical_calibration_mode": "unit_targets", + "apical_seed": null, + "augment_train": 1, + "batch_size": 128, + "bn_eps": 1e-05, + "bn_momentum": 0.1, + "data_dir": "/home/yurenh2/sdrn/data", + "depth": 20, + "device": "cuda", + "epochs": 20, + "eta_A": 0.01, + "eta_P": 0.01, + "eval_every": 0, + "eval_split": "validation", + "learn_P": 0, + "loader_seed": 0, + "lr": 0.03, + "lr_gamma": 0.1, + "lr_milestones": "100,150", + "lr_schedule": "cosine", + "max_steps": 0, + "mode": "hfa", + "momentum": 0.9, + "normalization": "batchnorm", + "nuisance_scale": 0.0, + "out": "results/oral_a_hfa_short/hfa_lr0.03.json", + "output_lr": 0.1, + "pert_directions": 1, + "pert_every": 4, + "pert_sigma": 0.01, + "perturb_seed": 1000, + "predictor_warmup_steps": 0, + "residual_scale": null, + "seed": 0, + "split_seed": 2027, + "train_limit": 10000, + "use_residual": 1, + "val_examples": 5000, + "vectorizer_mode": "spatial_template", + "warmup_epochs": 0, + "weight_decay": 0.0001, + "weight_scale": 1.0, + "width": 16 + }, + "calibration_metric_space": null, + "counters": { + "apical_warmup_examples": 0, + "calibration_event_examples": 0, + "causal_scalar_observations": 0, + "logical_batch_loss_queries": 0, + "ordinary_examples": 200000, + "per_example_loss_terms": 0, + "perturbation_events": 0, + "perturbation_forward_examples": 0, + "predictor_warmup_examples": 0 + }, + "diagnostics": { + "early_third_mean": 0.01133825796811531, + "innovation_negative_gradient_cosine": [ + 0.005747726187109947, + 0.003406030824407935, + 0.01511126384139061, + 0.0035506836138665676, + 0.038720425218343735, + 0.0014934181235730648, + 0.07037845999002457, + 0.010694023221731186, + 0.11927632242441177, + 0.029549002647399902, + 0.18039162456989288, + 0.023218173533678055, + 0.2376384288072586, + 0.039402417838573456, + 0.30495548248291016, + 0.2014918029308319, + 0.38542941212654114, + 0.2127426713705063, + 0.4092516303062439 + ], + "normalization_state": "training_batch_stats_without_running_update", + "raw_negative_gradient_cosine": [ + 0.005747726187109947, + 0.003406030824407935, + 0.01511126384139061, + 0.0035506836138665676, + 0.038720425218343735, + 0.0014934181235730648, + 0.07037845999002457, + 0.010694023221731186, + 0.11927632242441177, + 0.029549002647399902, + 0.18039162456989288, + 0.023218173533678055, + 0.2376384288072586, + 0.039402417838573456, + 0.30495548248291016, + 0.2014918029308319, + 0.38542941212654114, + 0.2127426713705063, + 0.4092516303062439 + ], + "teaching_negative_gradient_cosine": [ + 0.005747726187109947, + 0.003406030824407935, + 0.01511126384139061, + 0.0035506836138665676, + 0.038720425218343735, + 0.0014934181235730648, + 0.07037845999002457, + 0.010694023221731186, + 0.11927632242441177, + 0.029549002647399902, + 0.18039162456989288, + 0.023218173533678055, + 0.2376384288072586, + 0.039402417838573456, + 0.30495548248291016, + 0.2014918029308319, + 0.38542941212654114, + 0.2127426713705063, + 0.4092516303062439 + ], + "wall_s": 0.13352656364440918 + }, + "epochs": [ + { + "epoch": 1, + "lr": 0.03, + "output_lr": 0.1, + "step": 79, + "train_examples": 10000, + "train_loss": 2.7139521350860596 + }, + { + "epoch": 2, + "lr": 0.029815325108927065, + "output_lr": 0.0993844170297569, + "step": 158, + "train_examples": 10000, + "train_loss": 2.6086353378295897 + }, + { + "epoch": 3, + "lr": 0.029265847744427302, + "output_lr": 0.09755282581475769, + "step": 237, + "train_examples": 10000, + "train_loss": 2.4976746765136717 + }, + { + "epoch": 4, + "lr": 0.028365097862825515, + "output_lr": 0.09455032620941839, + "step": 316, + "train_examples": 10000, + "train_loss": 2.4073143131256103 + }, + { + "epoch": 5, + "lr": 0.02713525491562421, + "output_lr": 0.09045084971874738, + "step": 395, + "train_examples": 10000, + "train_loss": 2.360751438522339 + }, + { + "epoch": 6, + "lr": 0.02560660171779821, + "output_lr": 0.08535533905932738, + "step": 474, + "train_examples": 10000, + "train_loss": 2.2522952722549436 + }, + { + "epoch": 7, + "lr": 0.023816778784387097, + "output_lr": 0.07938926261462366, + "step": 553, + "train_examples": 10000, + "train_loss": 2.279150248336792 + }, + { + "epoch": 8, + "lr": 0.0218098574960932, + "output_lr": 0.07269952498697733, + "step": 632, + "train_examples": 10000, + "train_loss": 2.2757634071350097 + }, + { + "epoch": 9, + "lr": 0.01963525491562421, + "output_lr": 0.06545084971874737, + "step": 711, + "train_examples": 10000, + "train_loss": 2.0739892147064207 + }, + { + "epoch": 10, + "lr": 0.017346516975603462, + "output_lr": 0.05782172325201155, + "step": 790, + "train_examples": 10000, + "train_loss": 1.8923955001831054 + }, + { + "epoch": 11, + "lr": 0.015, + "output_lr": 0.05, + "step": 869, + "train_examples": 10000, + "train_loss": 1.9085709817886352 + }, + { + "epoch": 12, + "lr": 0.012653483024396533, + "output_lr": 0.04217827674798845, + "step": 948, + "train_examples": 10000, + "train_loss": 1.8094817079544068 + }, + { + "epoch": 13, + "lr": 0.01036474508437579, + "output_lr": 0.03454915028125263, + "step": 1027, + "train_examples": 10000, + "train_loss": 1.7567100723266602 + }, + { + "epoch": 14, + "lr": 0.0081901425039068, + "output_lr": 0.027300475013022664, + "step": 1106, + "train_examples": 10000, + "train_loss": 1.7179550045013428 + }, + { + "epoch": 15, + "lr": 0.006183221215612904, + "output_lr": 0.02061073738537635, + "step": 1185, + "train_examples": 10000, + "train_loss": 1.6542307502746583 + }, + { + "epoch": 16, + "lr": 0.004393398282201788, + "output_lr": 0.014644660940672627, + "step": 1264, + "train_examples": 10000, + "train_loss": 1.6502908784866333 + }, + { + "epoch": 17, + "lr": 0.00286474508437579, + "output_lr": 0.009549150281252633, + "step": 1343, + "train_examples": 10000, + "train_loss": 1.6148505542755127 + }, + { + "epoch": 18, + "lr": 0.001634902137174483, + "output_lr": 0.005449673790581611, + "step": 1422, + "train_examples": 10000, + "train_loss": 1.5908274923324586 + }, + { + "epoch": 19, + "lr": 0.000734152255572697, + "output_lr": 0.0024471741852423235, + "step": 1501, + "train_examples": 10000, + "train_loss": 1.570961207962036 + }, + { + "epoch": 20, + "lr": 0.0001846748910729351, + "output_lr": 0.0006155829702431171, + "step": 1580, + "train_examples": 10000, + "train_loss": 1.5774935535430907 + } + ], + "evaluation_protocol": { + "test_evaluations": 0, + "test_used_for_selection": false, + "validation_evaluations": 1 + }, + "final": { + "accuracy": 0.4158, + "epoch": 20, + "evaluation_split": "validation", + "finite": true, + "loss": 1.6006802001953124, + "step": 1580 + }, + "hardware": { + "cuda_device_name": "NVIDIA GeForce GTX 1080", + "cuda_visible_devices": "7", + "device": "cuda", + "device_total_memory_bytes": 8507949056, + "peak_memory_allocated_bytes": 1730831872, + "peak_memory_reserved_bytes": 2390753280, + "torch_version": "2.3.1+cu118" + }, + "protocol_family": "oral_a_cifar_local_resnet_development", + "provenance": { + "git_commit": "f37644eb452f0eb3d7f368740f8b39493f11b2aa", + "git_tracked_dirty": false + }, + "schema_version": 1, + "split": { + "cifar_source_files": [ + { + "bytes": 31035704, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_1", + "sha256": "54636561a3ce25bd3e19253c6b0d8538147b0ae398331ac4a2d86c6d987368cd" + }, + { + "bytes": 31035320, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_2", + "sha256": "766b2cef9fbc745cf056b3152224f7cf77163b330ea9a15f9392beb8b89bc5a8" + }, + { + "bytes": 31035999, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_3", + "sha256": "0f00d98ebfb30b3ec0ad19f9756dc2630b89003e10525f5e148445e82aa6a1f9" + }, + { + "bytes": 31035696, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_4", + "sha256": "3f7bb240661948b8f4d53e36ec720d8306f5668bd0071dcb4e6c947f78e9682b" + }, + { + "bytes": 31035623, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_5", + "sha256": "d91802434d8376bbaeeadf58a737e3a1b12ac839077e931237e0dcd43adcb154" + }, + { + "bytes": 31035526, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/test_batch", + "sha256": "f53d8d457504f7cff4ea9e021afcf0e0ad8e24a91f3fc42091b8adef61157831" + } + ], + "dataset": "cifar10", + "evaluation_split": "validation", + "input_layout": "NCHW", + "input_shape": [ + 3, + 32, + 32 + ], + "loader_seed": 0, + "normalization_mean": [ + 0.49140000343322754, + 0.4821999967098236, + 0.4465000033378601 + ], + "normalization_std": [ + 0.24699999392032623, + 0.2434999942779541, + 0.26159998774528503 + ], + "split_from_training_only": true, + "split_seed": 2027, + "test_examples": 10000, + "train_examples": 10000, + "training_augmentation": "random_crop_32_padding_4_zero_then_horizontal_flip_p0.5", + "validation_class_counts": { + "0": 500, + "1": 500, + "2": 500, + "3": 500, + "4": 500, + "5": 500, + "6": 500, + "7": 500, + "8": 500, + "9": 500 + }, + "validation_examples": 5000, + "validation_index_sha256": "8328b206a97c420e49e54e3eca4abe3274c4756b084355784ea3fb8059e4515b" + }, + "timing": { + "apical_warmup_wall_s": 0.0, + "evaluation_wall_s": 0.362001895904541, + "predictor_warmup_wall_s": 0.0, + "timing_excludes_data_loading_hashing_and_model_construction": true, + "total_timed_wall_s": 89.29601311683655, + "train_wall_s": 88.93181085586548 + }, + "work": { + "apical_macs_per_example": 40108672, + "causal_scalar_observations": 0, + "components": { + "apical_projection_macs": 8021734400000, + "apical_regression_macs": 0, + "bp_reverse_macs_estimate": 0, + "local_weight_correlation_macs": 8110208000000, + "ordinary_forward_macs": 8110208000000, + "perturbation_forward_macs": 0, + "warmup_clean_forward_macs": 0 + }, + "definition": "multiply-accumulates in conv/linear maps; one local weight correlation equals one forward-weight MAC count; BP reverse is estimated as one weight-gradient plus one activation-gradient convolution per forward convolution; elementwise nonlinearities and optimizer arithmetic excluded", + "forward_macs_per_example": 40551040, + "logical_batch_loss_queries": 0, + "per_example_cross_entropy_terms": 0, + "total_clean_forward_examples": 200000, + "total_forward_equivalent_examples": 200000, + "total_macs_estimate": 24242150400000 + } +} diff --git a/results/oral_a_hfa_short/hfa_lr0.1.json b/results/oral_a_hfa_short/hfa_lr0.1.json new file mode 100644 index 0000000..6bcb4ba --- /dev/null +++ b/results/oral_a_hfa_short/hfa_lr0.1.json @@ -0,0 +1,534 @@ +{ + "architecture": { + "adaptive_apical_parameters": 0, + "base_width": 16, + "blocks_per_stage": 3, + "bn_eps": 1e-05, + "bn_momentum": 0.1, + "depth": 20, + "family": "CIFAR 6n+2 ResNet, option-A shortcuts", + "fixed_feedback_parameters": 267904, + "fixed_traffic_coefficients": 0, + "forward_parameters": 269722, + "hidden_shapes": [ + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 16, + 32, + 32 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 32, + 16, + 16 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ], + [ + 64, + 8, + 8 + ] + ], + "normalization": "batchnorm", + "predictor_parameters": 0, + "residual_scale": 1.0, + "vectorizer_mode": null, + "vectorizer_parameters": 0 + }, + "args": { + "a_scale": 1.0, + "a_warmup_steps": 0, + "alignment_probe": 32, + "apical_calibration_mode": "unit_targets", + "apical_seed": null, + "augment_train": 1, + "batch_size": 128, + "bn_eps": 1e-05, + "bn_momentum": 0.1, + "data_dir": "/home/yurenh2/sdrn/data", + "depth": 20, + "device": "cuda", + "epochs": 20, + "eta_A": 0.01, + "eta_P": 0.01, + "eval_every": 0, + "eval_split": "validation", + "learn_P": 0, + "loader_seed": 0, + "lr": 0.1, + "lr_gamma": 0.1, + "lr_milestones": "100,150", + "lr_schedule": "cosine", + "max_steps": 0, + "mode": "hfa", + "momentum": 0.9, + "normalization": "batchnorm", + "nuisance_scale": 0.0, + "out": "results/oral_a_hfa_short/hfa_lr0.1.json", + "output_lr": 0.1, + "pert_directions": 1, + "pert_every": 4, + "pert_sigma": 0.01, + "perturb_seed": 1000, + "predictor_warmup_steps": 0, + "residual_scale": null, + "seed": 0, + "split_seed": 2027, + "train_limit": 10000, + "use_residual": 1, + "val_examples": 5000, + "vectorizer_mode": "spatial_template", + "warmup_epochs": 0, + "weight_decay": 0.0001, + "weight_scale": 1.0, + "width": 16 + }, + "calibration_metric_space": null, + "counters": { + "apical_warmup_examples": 0, + "calibration_event_examples": 0, + "causal_scalar_observations": 0, + "logical_batch_loss_queries": 0, + "ordinary_examples": 200000, + "per_example_loss_terms": 0, + "perturbation_events": 0, + "perturbation_forward_examples": 0, + "predictor_warmup_examples": 0 + }, + "diagnostics": { + "early_third_mean": 0.040405964168409504, + "innovation_negative_gradient_cosine": [ + 0.023671559989452362, + 0.013195669278502464, + 0.06748953461647034, + 0.017365306615829468, + 0.11135254800319672, + 0.009361166507005692, + 0.13842418789863586, + 0.01869494654238224, + 0.19049662351608276, + 0.023486969992518425, + 0.22018298506736755, + 0.033551692962646484, + 0.23453857004642487, + 0.03822024166584015, + 0.2642383873462677, + 0.18111935257911682, + 0.26726263761520386, + 0.16317608952522278, + 0.2346772849559784 + ], + "normalization_state": "training_batch_stats_without_running_update", + "raw_negative_gradient_cosine": [ + 0.023671559989452362, + 0.013195669278502464, + 0.06748953461647034, + 0.017365306615829468, + 0.11135254800319672, + 0.009361166507005692, + 0.13842418789863586, + 0.01869494654238224, + 0.19049662351608276, + 0.023486969992518425, + 0.22018298506736755, + 0.033551692962646484, + 0.23453857004642487, + 0.03822024166584015, + 0.2642383873462677, + 0.18111935257911682, + 0.26726263761520386, + 0.16317608952522278, + 0.2346772849559784 + ], + "teaching_negative_gradient_cosine": [ + 0.023671559989452362, + 0.013195669278502464, + 0.06748953461647034, + 0.017365306615829468, + 0.11135254800319672, + 0.009361166507005692, + 0.13842418789863586, + 0.01869494654238224, + 0.19049662351608276, + 0.023486969992518425, + 0.22018298506736755, + 0.033551692962646484, + 0.23453857004642487, + 0.03822024166584015, + 0.2642383873462677, + 0.18111935257911682, + 0.26726263761520386, + 0.16317608952522278, + 0.2346772849559784 + ], + "wall_s": 0.12181758880615234 + }, + "epochs": [ + { + "epoch": 1, + "lr": 0.1, + "output_lr": 0.1, + "step": 79, + "train_examples": 10000, + "train_loss": 3.256435613632202 + }, + { + "epoch": 2, + "lr": 0.0993844170297569, + "output_lr": 0.0993844170297569, + "step": 158, + "train_examples": 10000, + "train_loss": 3.5897572273254394 + }, + { + "epoch": 3, + "lr": 0.09755282581475769, + "output_lr": 0.09755282581475769, + "step": 237, + "train_examples": 10000, + "train_loss": 3.483150173187256 + }, + { + "epoch": 4, + "lr": 0.09455032620941839, + "output_lr": 0.09455032620941839, + "step": 316, + "train_examples": 10000, + "train_loss": 3.499296664428711 + }, + { + "epoch": 5, + "lr": 0.09045084971874738, + "output_lr": 0.09045084971874738, + "step": 395, + "train_examples": 10000, + "train_loss": 3.5787659603118898 + }, + { + "epoch": 6, + "lr": 0.08535533905932738, + "output_lr": 0.08535533905932738, + "step": 474, + "train_examples": 10000, + "train_loss": 2.639942427062988 + }, + { + "epoch": 7, + "lr": 0.07938926261462366, + "output_lr": 0.07938926261462366, + "step": 553, + "train_examples": 10000, + "train_loss": 3.3110560848236084 + }, + { + "epoch": 8, + "lr": 0.07269952498697733, + "output_lr": 0.07269952498697733, + "step": 632, + "train_examples": 10000, + "train_loss": 2.8327583209991456 + }, + { + "epoch": 9, + "lr": 0.06545084971874737, + "output_lr": 0.06545084971874737, + "step": 711, + "train_examples": 10000, + "train_loss": 2.2873985042572023 + }, + { + "epoch": 10, + "lr": 0.05782172325201155, + "output_lr": 0.05782172325201155, + "step": 790, + "train_examples": 10000, + "train_loss": 2.1511593353271485 + }, + { + "epoch": 11, + "lr": 0.05, + "output_lr": 0.05, + "step": 869, + "train_examples": 10000, + "train_loss": 2.080998848724365 + }, + { + "epoch": 12, + "lr": 0.04217827674798845, + "output_lr": 0.04217827674798845, + "step": 948, + "train_examples": 10000, + "train_loss": 1.880720576095581 + }, + { + "epoch": 13, + "lr": 0.03454915028125263, + "output_lr": 0.03454915028125263, + "step": 1027, + "train_examples": 10000, + "train_loss": 1.80372526512146 + }, + { + "epoch": 14, + "lr": 0.027300475013022664, + "output_lr": 0.027300475013022664, + "step": 1106, + "train_examples": 10000, + "train_loss": 1.7575418134689331 + }, + { + "epoch": 15, + "lr": 0.02061073738537635, + "output_lr": 0.02061073738537635, + "step": 1185, + "train_examples": 10000, + "train_loss": 1.6703623008728028 + }, + { + "epoch": 16, + "lr": 0.014644660940672627, + "output_lr": 0.014644660940672627, + "step": 1264, + "train_examples": 10000, + "train_loss": 1.6365677841186523 + }, + { + "epoch": 17, + "lr": 0.009549150281252633, + "output_lr": 0.009549150281252633, + "step": 1343, + "train_examples": 10000, + "train_loss": 1.6003902708053588 + }, + { + "epoch": 18, + "lr": 0.005449673790581611, + "output_lr": 0.005449673790581611, + "step": 1422, + "train_examples": 10000, + "train_loss": 1.5677705976486207 + }, + { + "epoch": 19, + "lr": 0.0024471741852423235, + "output_lr": 0.0024471741852423235, + "step": 1501, + "train_examples": 10000, + "train_loss": 1.5354159210205078 + }, + { + "epoch": 20, + "lr": 0.0006155829702431171, + "output_lr": 0.0006155829702431171, + "step": 1580, + "train_examples": 10000, + "train_loss": 1.5316140928268434 + } + ], + "evaluation_protocol": { + "test_evaluations": 0, + "test_used_for_selection": false, + "validation_evaluations": 1 + }, + "final": { + "accuracy": 0.4352, + "epoch": 20, + "evaluation_split": "validation", + "finite": true, + "loss": 1.5658135009765626, + "step": 1580 + }, + "hardware": { + "cuda_device_name": "NVIDIA GeForce GTX 1080", + "cuda_visible_devices": "5", + "device": "cuda", + "device_total_memory_bytes": 8507949056, + "peak_memory_allocated_bytes": 1730831872, + "peak_memory_reserved_bytes": 2390753280, + "torch_version": "2.3.1+cu118" + }, + "protocol_family": "oral_a_cifar_local_resnet_development", + "provenance": { + "git_commit": "f37644eb452f0eb3d7f368740f8b39493f11b2aa", + "git_tracked_dirty": false + }, + "schema_version": 1, + "split": { + "cifar_source_files": [ + { + "bytes": 31035704, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_1", + "sha256": "54636561a3ce25bd3e19253c6b0d8538147b0ae398331ac4a2d86c6d987368cd" + }, + { + "bytes": 31035320, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_2", + "sha256": "766b2cef9fbc745cf056b3152224f7cf77163b330ea9a15f9392beb8b89bc5a8" + }, + { + "bytes": 31035999, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_3", + "sha256": "0f00d98ebfb30b3ec0ad19f9756dc2630b89003e10525f5e148445e82aa6a1f9" + }, + { + "bytes": 31035696, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_4", + "sha256": "3f7bb240661948b8f4d53e36ec720d8306f5668bd0071dcb4e6c947f78e9682b" + }, + { + "bytes": 31035623, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/data_batch_5", + "sha256": "d91802434d8376bbaeeadf58a737e3a1b12ac839077e931237e0dcd43adcb154" + }, + { + "bytes": 31035526, + "path": "/home/yurenh2/sdrn/data/cifar-10-batches-py/test_batch", + "sha256": "f53d8d457504f7cff4ea9e021afcf0e0ad8e24a91f3fc42091b8adef61157831" + } + ], + "dataset": "cifar10", + "evaluation_split": "validation", + "input_layout": "NCHW", + "input_shape": [ + 3, + 32, + 32 + ], + "loader_seed": 0, + "normalization_mean": [ + 0.49140000343322754, + 0.4821999967098236, + 0.4465000033378601 + ], + "normalization_std": [ + 0.24699999392032623, + 0.2434999942779541, + 0.26159998774528503 + ], + "split_from_training_only": true, + "split_seed": 2027, + "test_examples": 10000, + "train_examples": 10000, + "training_augmentation": "random_crop_32_padding_4_zero_then_horizontal_flip_p0.5", + "validation_class_counts": { + "0": 500, + "1": 500, + "2": 500, + "3": 500, + "4": 500, + "5": 500, + "6": 500, + "7": 500, + "8": 500, + "9": 500 + }, + "validation_examples": 5000, + "validation_index_sha256": "8328b206a97c420e49e54e3eca4abe3274c4756b084355784ea3fb8059e4515b" + }, + "timing": { + "apical_warmup_wall_s": 0.0, + "evaluation_wall_s": 0.3655712604522705, + "predictor_warmup_wall_s": 0.0, + "timing_excludes_data_loading_hashing_and_model_construction": true, + "total_timed_wall_s": 89.62506937980652, + "train_wall_s": 89.25713443756104 + }, + "work": { + "apical_macs_per_example": 40108672, + "causal_scalar_observations": 0, + "components": { + "apical_projection_macs": 8021734400000, + "apical_regression_macs": 0, + "bp_reverse_macs_estimate": 0, + "local_weight_correlation_macs": 8110208000000, + "ordinary_forward_macs": 8110208000000, + "perturbation_forward_macs": 0, + "warmup_clean_forward_macs": 0 + }, + "definition": "multiply-accumulates in conv/linear maps; one local weight correlation equals one forward-weight MAC count; BP reverse is estimated as one weight-gradient plus one activation-gradient convolution per forward convolution; elementwise nonlinearities and optimizer arithmetic excluded", + "forward_macs_per_example": 40551040, + "logical_batch_loss_queries": 0, + "per_example_cross_entropy_terms": 0, + "total_clean_forward_examples": 200000, + "total_forward_equivalent_examples": 200000, + "total_macs_estimate": 24242150400000 + } +} diff --git a/results/oral_a_hfa_short_selection.json b/results/oral_a_hfa_short_selection.json new file mode 100644 index 0000000..394de01 --- /dev/null +++ b/results/oral_a_hfa_short_selection.json @@ -0,0 +1,69 @@ +{ + "comparators": { + "matched_short_bp_accuracy": 0.7494, + "matched_short_dfa_accuracy": 0.3716, + "matched_short_failed_v1_sdil_accuracy": 0.4198 + }, + "confirmation_test_seeds_touched": false, + "full_validation_threshold": 0.5, + "protocol": "convolutional_hfa_S1_v1", + "rows": [ + { + "accuracy": 0.3964, + "early_third_alignment": 0.004572224900281678, + "finite": true, + "fixed_feedback_parameters": 267904, + "logical_batch_loss_queries": 0, + "loss": 1.6798031494140624, + "lr": 0.01, + "path": "results/oral_a_hfa_short/hfa_lr0.01.json", + "peak_memory_allocated_bytes": 1730831872, + "source_commit": "f37644eb452f0eb3d7f368740f8b39493f11b2aa", + "total_macs": 24242150400000, + "wall_s": 89.53109097480774 + }, + { + "accuracy": 0.4158, + "early_third_alignment": 0.01133825796811531, + "finite": true, + "fixed_feedback_parameters": 267904, + "logical_batch_loss_queries": 0, + "loss": 1.6006802001953124, + "lr": 0.03, + "path": "results/oral_a_hfa_short/hfa_lr0.03.json", + "peak_memory_allocated_bytes": 1730831872, + "source_commit": "f37644eb452f0eb3d7f368740f8b39493f11b2aa", + "total_macs": 24242150400000, + "wall_s": 89.29601311683655 + }, + { + "accuracy": 0.4352, + "early_third_alignment": 0.040405964168409504, + "finite": true, + "fixed_feedback_parameters": 267904, + "logical_batch_loss_queries": 0, + "loss": 1.5658135009765626, + "lr": 0.1, + "path": "results/oral_a_hfa_short/hfa_lr0.1.json", + "peak_memory_allocated_bytes": 1730831872, + "source_commit": "f37644eb452f0eb3d7f368740f8b39493f11b2aa", + "total_macs": 24242150400000, + "wall_s": 89.62506937980652 + } + ], + "selected": { + "accuracy": 0.4352, + "early_third_alignment": 0.040405964168409504, + "finite": true, + "fixed_feedback_parameters": 267904, + "logical_batch_loss_queries": 0, + "loss": 1.5658135009765626, + "lr": 0.1, + "path": "results/oral_a_hfa_short/hfa_lr0.1.json", + "peak_memory_allocated_bytes": 1730831872, + "source_commit": "f37644eb452f0eb3d7f368740f8b39493f11b2aa", + "total_macs": 24242150400000, + "wall_s": 89.62506937980652 + }, + "status": "selected_full_closed" +} -- cgit v1.2.3