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": 3, "device": "cuda", "outdir": "results", "tag": "ep_original_v2_mnist_ep_w500_d1_s3"}, "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": "887ecfb29478e9a7456a12ea85f7ba7448e7277d", "git_dirty": false}, "steps": [{"epoch_end": 0, "test_acc": 0.9541, "test_loss": 0.07772458267211914, "train_loss": 0.00803004764020443}, {"epoch_end": 1, "test_acc": 0.965, "test_loss": 0.059839314651489256, "train_loss": 0.005804809276014566}, {"epoch_end": 2, "test_acc": 0.9676, "test_loss": 0.05486894283294678, "train_loss": 0.002006430411711335}, {"epoch_end": 3, "test_acc": 0.9696, "test_loss": 0.05063978977203369, "train_loss": 0.0010454378789290786}, {"epoch_end": 4, "test_acc": 0.9706, "test_loss": 0.049372467613220214, "train_loss": 0.0}, {"epoch_end": 5, "test_acc": 0.9704, "test_loss": 0.05071136360168457, "train_loss": 1.2589122206918546e-06}, {"epoch_end": 6, "test_acc": 0.9715, "test_loss": 0.04733491535186768, "train_loss": 0.0}, {"epoch_end": 7, "test_acc": 0.9713, "test_loss": 0.0485486837387085, "train_loss": 0.0}, {"epoch_end": 8, "test_acc": 0.9701, "test_loss": 0.052020733451843264, "train_loss": 0.0}, {"epoch_end": 9, "test_acc": 0.9648, "test_loss": 0.05864917869567871, "train_loss": 0.0}, {"epoch_end": 10, "test_acc": 0.9656, "test_loss": 0.05929891471862793, "train_loss": 0.0}, {"epoch_end": 11, "test_acc": 0.9603, "test_loss": 0.06528462982177734, "train_loss": 0.0}, {"epoch_end": 12, "test_acc": 0.9695, "test_loss": 0.05300445156097412, "train_loss": 2.462031079630833e-05}, {"epoch_end": 13, "test_acc": 0.9698, "test_loss": 0.05086651725769043, "train_loss": 0.0}, {"epoch_end": 14, "test_acc": 0.9658, "test_loss": 0.058786696243286134, "train_loss": 0.0}, {"epoch_end": 15, "test_acc": 0.9598, "test_loss": 0.06823719139099121, "train_loss": 0.0}, {"epoch_end": 16, "test_acc": 0.9667, "test_loss": 0.057434351921081546, "train_loss": 0.0}, {"epoch_end": 17, "test_acc": 0.9701, "test_loss": 0.05087097434997558, "train_loss": 0.0}, {"epoch_end": 18, "test_acc": 0.9707, "test_loss": 0.050260302734375, "train_loss": 0.0}, {"epoch_end": 19, "test_acc": 0.9702, "test_loss": 0.049360933685302734, "train_loss": 0.0}, {"epoch_end": 20, "test_acc": 0.9706, "test_loss": 0.05029971561431885, "train_loss": 0.0}, {"epoch_end": 21, "test_acc": 0.9708, "test_loss": 0.048192334842681886, "train_loss": 0.0}, {"epoch_end": 22, "test_acc": 0.9705, "test_loss": 0.04882320232391357, "train_loss": 0.0}, {"epoch_end": 23, "test_acc": 0.9708, "test_loss": 0.0485589602470398, "train_loss": 0.0}, {"epoch_end": 24, "test_acc": 0.9709, "test_loss": 0.048532071304321286, "train_loss": 0.0}], "final": {"test_acc": 0.9709, "test_loss": 0.048532071304321286, "wall_s": 772.848649263382}}
|
