1
|
{"args": {"method": "ep", "dataset": "mnist", "depth": 1, "width": 500, "epochs": 25, "batch_size": 20, "train_examples": 50000, "max_batches": 0, "eta": null, "momentum": 0.9, "act": "tanh", "residual": 0, "feedback_scale": 1.0, "pepita_f_scale": 0.05, "pepita_decay_epochs": null, "keep_prob": 0.9, "ff_threshold": 2.0, "ep_beta": null, "ep_dt": 0.5, "ep_free_steps": null, "ep_nudge_steps": null, "ep_persistent": 1, "canonical_preprocess": 1, "probe_bs": 512, "seed": 1, "device": "cuda", "outdir": "results", "tag": "ep_original_v2_mnist_ep_w500_d1_s1"}, "resolved_protocol": {"beta": 0.5, "free_steps": 20, "nudge_steps": 4, "canonical_depth": true, "learning_rates": [0.1, 0.05], "persistent_particles": true, "train_examples": 50000}, "method_source": {"paper": "https://www.frontiersin.org/journals/computational-neuroscience/articles/10.3389/fncom.2017.00024/full", "code": "https://github.com/bscellier/Towards-a-Biologically-Plausible-Backprop", "protocol": "two-phase energy-gradient dynamics; persistent free particles; random beta sign"}, "provenance": {"git_commit": "5b9de3f36c91ea8ad57fc4dae5aa0c04f08a7a26", "git_dirty": false}, "steps": [{"epoch_end": 0, "test_acc": 0.955, "test_loss": 0.07901275215148926, "train_loss": 0.01013753842562437}, {"epoch_end": 1, "test_acc": 0.9645, "test_loss": 0.06253083763122559, "train_loss": 0.005079905968159437}, {"epoch_end": 2, "test_acc": 0.966, "test_loss": 0.058018816566467284, "train_loss": 0.0012329824967309833}, {"epoch_end": 3, "test_acc": 0.9684, "test_loss": 0.05564236717224121, "train_loss": 0.0008007634314708412}, {"epoch_end": 4, "test_acc": 0.9657, "test_loss": 0.059846015548706055, "train_loss": 0.0004937855992466211}, {"epoch_end": 5, "test_acc": 0.9684, "test_loss": 0.05555516471862793, "train_loss": 0.0007326625636778772}, {"epoch_end": 6, "test_acc": 0.9702, "test_loss": 0.05117148704528809, "train_loss": 0.00024014394148252904}, {"epoch_end": 7, "test_acc": 0.9687, "test_loss": 0.054611382865905764, "train_loss": 4.7521469241473824e-05}, {"epoch_end": 8, "test_acc": 0.9627, "test_loss": 0.06397499732971192, "train_loss": 0.000436745845945552}, {"epoch_end": 9, "test_acc": 0.9685, "test_loss": 0.05292844734191895, "train_loss": 6.424915568459255e-07}, {"epoch_end": 10, "test_acc": 0.9688, "test_loss": 0.052844040870666506, "train_loss": 1.239909437344977e-07}, {"epoch_end": 11, "test_acc": 0.9653, "test_loss": 0.05899420146942139, "train_loss": 0.0}, {"epoch_end": 12, "test_acc": 0.9667, "test_loss": 0.05708587989807129, "train_loss": 0.0}, {"epoch_end": 13, "test_acc": 0.965, "test_loss": 0.059490954399108883, "train_loss": 0.0}, {"epoch_end": 14, "test_acc": 0.9633, "test_loss": 0.0638974765777588, "train_loss": 0.0}, {"epoch_end": 15, "test_acc": 0.9681, "test_loss": 0.05465109443664551, "train_loss": 0.0005773885641247034}, {"epoch_end": 16, "test_acc": 0.9667, "test_loss": 0.05632556095123291, "train_loss": 0.0008614619146101177}, {"epoch_end": 17, "test_acc": 0.9671, "test_loss": 0.05757739877700806, "train_loss": 0.0}, {"epoch_end": 18, "test_acc": 0.9625, "test_loss": 0.06238394584655762, "train_loss": 0.0}, {"epoch_end": 19, "test_acc": 0.9675, "test_loss": 0.05585011367797851, "train_loss": 0.0}, {"epoch_end": 20, "test_acc": 0.9687, "test_loss": 0.053836871910095216, "train_loss": 0.0}, {"epoch_end": 21, "test_acc": 0.9689, "test_loss": 0.051564078330993655, "train_loss": 0.0}, {"epoch_end": 22, "test_acc": 0.967, "test_loss": 0.054196818923950195, "train_loss": 0.0}, {"epoch_end": 23, "test_acc": 0.9673, "test_loss": 0.05569706926345825, "train_loss": 0.0}, {"epoch_end": 24, "test_acc": 0.9685, "test_loss": 0.051756802940368656, "train_loss": 0.0}], "final": {"test_acc": 0.9685, "test_loss": 0.051756802940368656, "wall_s": 784.3743386268616}}
|
