Split (2)

train · 50 rows

data_source stringclasses 1 value	prompt listlengths 1 1	ability stringclasses 1 value	reward_model dict	extra_info dict
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nBars problem for RBM training -- the canonical Hinton 2000 / Foldiak 1990\nsanity check for contrastive divergence.\n\nProblem:\n Generate 4x4 binary images by independently activating each of 8 possible\n bars (4 horizontal rows + 4 vertical columns) with a small prob...	{ "slug": "bars-rbm", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/bars-rbm/bars_rbm.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nMNIST-2x5 subclass distillation.\n\nReproduction of the central experiment from Mueller, Kornblith & Hinton,\n\"Subclass distillation\", arXiv:2002.03936 (2020).\n\nSetup\n-----\n- Re-label MNIST: digits 0..4 -> super-class A, digits 5..9 -> super-class B.\n- The TEACHER...	{ "slug": "mnist-2x5-subclass", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/mnist-2x5-subclass/mnist_2x5_subclass.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nGrapheme-sememe synthetic word reading + lesion / relearning experiments.\n\nSource:\n Hinton & Sejnowski (1986), \"Learning and relearning in Boltzmann machines\",\n in Rumelhart & McClelland (eds.), Parallel Distributed Processing, Vol. 1,\n Chapter 7.\n\nPr...	{ "slug": "grapheme-sememe", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/grapheme-sememe/grapheme_sememe.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nFast weights to deblur old memories (Hinton & Plaut 1987).\n\nSource:\n G. E. Hinton and D. C. Plaut (1987), \"Using Fast Weights to Deblur Old\n Memories\", Proceedings of the Ninth Annual Conference of the Cognitive\n Science Society, pp. 177-186.\n\nProblem:\...	{ "slug": "fast-weights-rehearsal", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/fast-weights-rehearsal/fast_weights_rehearsal.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nHelmholtz-machine shifter (Dayan, Hinton, Neal & Zemel 1995, \"The Helmholtz\nmachine\", Neural Computation 7(5):889--904).\n\nProblem:\n 4 x 8 binary image. Row 0 (and its duplicate row 1) is uniform-Bernoulli with\n marginal p_on=0.2. Rows 2-3 (duplicates) are row 0 ...	{ "slug": "helmholtz-shifter", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/helmholtz-shifter/helmholtz_shifter.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\n8-3-8 backprop autoencoder -- backprop mirror of Ackley/Hinton/Sejnowski (1985)\nencoder, trained the way Rumelhart, Hinton & Williams (1986) trained their\nMLPs in \"Learning internal representations by error propagation\" (PDP Vol. 1,\nCh. 8).\n\nProblem:\n 8 one-hot ...	{ "slug": "encoder-backprop-8-3-8", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/encoder-backprop-8-3-8/encoder_backprop_8_3_8.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\n4-3-4 encoder -- over-complete Boltzmann-machine reproduction from\nAckley, Hinton & Sejnowski, \"A learning algorithm for Boltzmann machines\",\nCognitive Science 9 (1985).\n\nProblem:\n Two groups of 4 visible binary units (V1, V2) connected through 3 hidden\n binary...	{ "slug": "encoder-4-3-4", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/encoder-4-3-4/encoder_4_3_4.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"AIR (Attend, Infer, Repeat) on Multi-MNIST scenes -- pure numpy.\n\nEslami, Heess, Weber, Tassa, Szepesvari, Kavukcuoglu & Hinton (NIPS 2016).\n\nThe full AIR model is recurrent: at each step t the model decides\n z_pres_t -- whether to attend to another object (Bernoul...	{ "slug": "air-multimnist", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/air-multimnist/air_multimnist.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\n4-2-4 encoder — Boltzmann-machine reproduction of the experiment from\nAckley, Hinton & Sejnowski, \"A learning algorithm for Boltzmann machines\",\nCognitive Science 9 (1985).\n\nProblem:\n Two groups of 4 visible binary units (V1, V2) connected through 2 hidden\n bin...	{ "slug": "encoder-4-2-4", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/encoder-4-2-4/encoder_4_2_4.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nBars task — Helmholtz machine + wake-sleep reproduction of the experiment from\nHinton, Dayan, Frey & Neal, \"The wake-sleep algorithm for unsupervised neural\nnetworks\", Science 268 (1995).\n\nProblem:\n 4x4 binary images. Each image has either vertical bars (with p...	{ "slug": "bars", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/bars/bars.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nAIR with a programmable 3D renderer (Eslami, Heess, Weber, Tassa, Szepesvari,\nKavukcuoglu & Hinton, 2016).\n\nPipeline\n--------\n1. A pure-numpy Lambertian renderer that ray-casts up to 3 unit primitives\n (sphere, cube, cylinder) under an orthographic camera and a s...	{ "slug": "air-3d-primitives", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/air-3d-primitives/air_3d_primitives.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nGated three-way conditional RBM trained on (x, y) pairs of binary\nrandom-dot images, where y is x transformed by a known transformation\ndrawn from {shifts of +/- k pixels, rotations of multiples of 90 deg}.\n\nReference: Memisevic & Hinton, \"Unsupervised learning of i...	{ "slug": "transforming-pairs", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/transforming-pairs/transforming_pairs.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nTransforming auto-encoders -- the seminal capsules paper.\nReproduction of Hinton, Krizhevsky & Wang, \"Transforming auto-encoders\",\nICANN 2011.\n\nTranslation-only variant: each \"capsule\" learns to output (presence_prob, x, y)\nfor its entity. The transformation ent...	{ "slug": "transforming-autoencoders", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/transforming-autoencoders/transforming_autoencoders.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nBouncing balls (2 balls, TRBM) — reproduction of the synthetic-video benchmark\nfrom Sutskever & Hinton, \"Learning multilevel distributed representations for\nhigh-dimensional sequences\", AISTATS 2007.\n\nProblem:\n A short video of two balls bouncing around in a bo...	{ "slug": "bouncing-balls-2", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/bouncing-balls-2/bouncing_balls_2.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\n40-10-40 encoder -- Boltzmann-machine reproduction of the experiment from\nAckley, Hinton & Sejnowski, \"A learning algorithm for Boltzmann machines\",\nCognitive Science 9 (1985).\n\nProblem:\n Two groups of 40 visible binary units (V1, V2) connected through 10 hidden\...	{ "slug": "encoder-40-10-40", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/encoder-40-10-40/encoder_40_10_40.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nRecurrent shift register (Rumelhart, Hinton & Williams 1986, PDP Vol. 1, Ch. 8).\n\nA small recurrent network with N hidden units learns to act as an N-stage\nshift register: random binary bits arrive one at a time on a single input\nline, and the network's job is to emi...	{ "slug": "recurrent-shift-register", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/recurrent-shift-register/recurrent_shift_register.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\n3-bit even-parity ensemble — Boltzmann-machine reproduction of the negative\nresult motivating the encoder problems in Ackley, Hinton & Sejnowski,\n\"A learning algorithm for Boltzmann machines\", Cognitive Science 9 (1985).\n\nProblem:\n Three visible binary units. Tra...	{ "slug": "encoder-3-parity", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/encoder-3-parity/encoder_3_parity.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nSpline images & factorial vector quantization — reproduction of Hinton & Zemel,\n\"Autoencoders, MDL and Helmholtz free energy\", NIPS 6 (1994).\n\nProblem\n-------\n200 images (8 x 12) formed by Gaussian-blurring a curve through 5 control\npoints. The y-position of each...	{ "slug": "spline-images-factorial-vq", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/spline-images-factorial-vq/spline_images_factorial_vq.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nMNIST knowledge distillation with the digit \"3\" omitted from the student's\ntraining set.\n\nSource: Hinton, Vinyals & Dean (2015), \"Distilling the knowledge in a neural\nnetwork\", NIPS Deep Learning Workshop. Section 3 reports that a student trained\non a transfer s...	{ "slug": "distillation-mnist-omitted-3", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/distillation-mnist-omitted-3/distillation_mnist_omitted_3.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\n2-bit distributed-to-local mapping with a 1-unit bottleneck\n(Rumelhart, Hinton & Williams 1986, PDP Vol. 1, Ch. 8).\n\nProblem:\n 4 distributed 2-bit inputs map to 4 one-hot 4-D output targets:\n (0, 0) -> [1, 0, 0, 0]\n (0, 1) -> [0, 1, 0, 0]\n ...	{ "slug": "distributed-to-local-bottleneck", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/distributed-to-local-bottleneck/distributed_to_local_bottleneck.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nFast-weights associative retrieval (Ba, Hinton, Mnih, Leibo, Ionescu 2016).\n\nSource:\n J. Ba, G. Hinton, V. Mnih, J. Z. Leibo, C. Ionescu (2016),\n \"Using Fast Weights to Attend to the Recent Past\", NIPS.\n https://arxiv.org/abs/1610.06258\n\nProblem:\n S...	{ "slug": "fast-weights-associative-retrieval", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/fast-weights-associative-retrieval/fast_weights_associative_retrieval.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nMulti-level glimpse MNIST (Ba, Hinton, Mnih, Leibo & Ionescu 2016).\n\nSource:\n J. Ba, G. Hinton, V. Mnih, J. Z. Leibo, C. Ionescu (2016),\n \"Using Fast Weights to Attend to the Recent Past\", NIPS.\n https://arxiv.org/abs/1610.06258\n\nProblem:\n The 28x28...	{ "slug": "multi-level-glimpse-mnist", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/multi-level-glimpse-mnist/multi_level_glimpse_mnist.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nForward-Forward unsupervised on MNIST with hybrid-image negatives\n(Hinton 2022, \"The forward-forward algorithm: some preliminary investigations\").\n\nEach layer is trained locally to push its goodness (sum of squared\npost-ReLU activations) UP for real digits and ...	{ "slug": "ff-hybrid-mnist", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/ff-hybrid-mnist/ff_hybrid_mnist.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\naffNIST robustness test (Sabour, Frosst & Hinton 2017, \"Dynamic routing between capsules\").\n\nTrain both a CapsNet and a parameter-matched CNN on translated MNIST (40x40\ncanvas, MNIST digit randomly placed within +-6 px of center). Test on affNIST\n40x40 (rea...	{ "slug": "affnist", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/affnist/affnist.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nCatch with delayed-blank observations (Ba, Hinton, Mnih, Leibo, Ionescu 2016).\n\nSource:\n J. Ba, G. Hinton, V. Mnih, J. Z. Leibo, C. Ionescu (2016),\n \"Using Fast Weights to Attend to the Recent Past\", NIPS.\n https://arxiv.org/abs/1610.06258 (section 5: \"...	{ "slug": "catch-game", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/catch-game/catch_game.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nT-C discrimination via a weight-tied (convolutional) network — early-CNN\nreproduction of Rumelhart, Hinton & Williams (1986), PDP Vol. 1, Ch. 8.\n\nProblem:\n 8 patterns: a \"block T\" and a \"block C\", each in 4 rotations (0, 90, 180,\n 270 degrees), drawn on a ...	{ "slug": "t-c-discrimination", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/t-c-discrimination/t_c_discrimination.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nSynthetic-spectrogram riser/non-riser discrimination\n(Plaut & Hinton 1987, \"Learning sets of filters using back-propagation\",\nComputer Speech and Language, vol 2, pp 35-61).\n\nA 6 frequency x 9 time = 54-D synthetic spectrogram with one \"track\"\n(a path of frequen...	{ "slug": "riser-spectrogram", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/riser-spectrogram/riser_spectrogram.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nAdaptive mixture of local experts on Peterson-Barney vowels.\n\nReplicates the headline experiment of Jacobs, Jordan, Nowlan & Hinton (1991),\n\"Adaptive mixtures of local experts\", Neural Computation 3(1):79-87.\n\nTask: 4-class speaker-independent vowel classification...	{ "slug": "vowel-mixture-experts", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/vowel-mixture-experts/vowel_mixture_experts.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"Forward-Forward, top-down recurrent on repeated-frame MNIST (Hinton 2022).\n\nA static MNIST digit is treated as a \"video\" of repeated identical frames.\nEach hidden layer at time t is computed from the L2-normalized activities of\nthe layer above and below at t-1. The t...	{ "slug": "ff-recurrent-mnist", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/ff-recurrent-mnist/ff_recurrent_mnist.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nFamily-trees / kinship task — Hinton (1986), \"Learning distributed\nrepresentations of concepts\", Cognitive Science 8.\n\nTwo isomorphic 12-person family trees (English + Italian) and 12 kinship\nrelations. The network learns to predict P2 from (P1, R). The 6-unit\n\"p...	{ "slug": "family-trees", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/family-trees/family_trees.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\n6-bit symmetry / palindrome detection (Rumelhart, Hinton & Williams 1986).\n\nA 6-2-1 sigmoid network learns to output 1 iff the 6-bit input is a\npalindrome (symmetric about its midpoint). All 64 patterns are enumerated;\n8 are palindromes.\n\nFamous result (RHW1986, Fi...	{ "slug": "symmetry", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/symmetry/symmetry.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nConstellations -- 2D point-cloud part-whole grouping.\n\nNumpy reproduction of Kosiorek, Sabour, Teh & Hinton, \"Stacked capsule\nautoencoders\" (NeurIPS 2019), constellations experiment.\n\nEach example is the union of K=3 affine-transformed copies of fixed point\ntempl...	{ "slug": "constellations", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/constellations/constellations.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nMultiMNIST + Capsule Network with Dynamic Routing.\n\nNumpy reproduction of Sabour, Frosst & Hinton, \"Dynamic routing between\ncapsules\", NIPS 2017. The point of the paper is that capsules separately\nidentify and reconstruct heavily overlapping digits via routing-by-a...	{ "slug": "multimnist-capsnet", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/multimnist-capsnet/multimnist_capsnet.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nHochreiter & Schmidhuber pathological long-term-dependency RNN tasks\n(Sutskever, Martens, Dahl & Hinton 2013, \"On the importance of\ninitialization and momentum in deep learning\", ICML).\n\nThe headline of Sutskever et al. 2013 is that vanilla RNNs trained with\nSGD +...	{ "slug": "rnn-pathological", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/rnn-pathological/rnn_pathological.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nsmallNORB held-out viewpoint generalization (Hinton, Sabour & Frosst 2018,\n\"Matrix capsules with EM routing\", ICLR).\n\nReproduces the headline claim: a capsule network with 4x4 pose matrices and\nEM routing extrapolates better to novel viewpoints than a parameter-m...	{ "slug": "smallnorb-novel-viewpoint", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/smallnorb-novel-viewpoint/smallnorb_novel_viewpoint.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nNegation problem (Rumelhart, Hinton & Williams 1986),\n\"Learning internal representations by error propagation\",\nPDP Vol. 1, Ch. 8, §\"Symmetry / Negation / Encoder examples\".\n\nProblem:\n 4 inputs (1 flag bit + 3 data bits) -> 3 outputs.\n 16 patterns total (...	{ "slug": "negation", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/negation/negation.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nDipole 3D-constraint population code (Zemel & Hinton 1995).\n\nGenerates 8x8 dipole images parameterized by (x, y, theta) -- three varying\nparameters -- trains an unsupervised population coder with 225 hidden units,\nand recovers a 3D implicit space whose coordinates co...	{ "slug": "dipole-3d-constraint", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/dipole-3d-constraint/dipole_3d_constraint.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nDeep Lambertian Spheres -- joint recovery of albedo + surface normals from\nmultiple lit views of synthetic spheres.\n\nSource:\n Tang, Salakhutdinov & Hinton (2012),\n \"Deep Lambertian Networks\", ICML.\n\nArchitecture (v1, simplified from the paper's GRBM):\n ...	{ "slug": "deep-lambertian-spheres", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/deep-lambertian-spheres/deep_lambertian_spheres.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nThree bouncing balls modeled with a Recurrent Temporal RBM (RTRBM).\n\nSource:\n Sutskever, I., Hinton, G. E., & Taylor, G. W. (2008/2009),\n \"The recurrent temporal restricted Boltzmann machine,\" NIPS 21.\n\nThe RTRBM is a TRBM whose hidden layer at time t-1 directl...	{ "slug": "bouncing-balls-3", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/bouncing-balls-3/bouncing_balls_3.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\n25-sequence look-up RNN task (Rumelhart, Hinton & Williams 1986, PDP Vol. 1, Ch. 8).\n\nA small recurrent network with 30 tanh hidden units learns to associate 25\nfive-letter sequences with their 3-bit target codes. Inputs are one-hot\nencoded letters from a 5-letter al...	{ "slug": "sequence-lookup-25", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/sequence-lookup-25/sequence_lookup_25.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nN-bit parity backprop, faithful to Rumelhart, Hinton & Williams (1986),\n\"Learning internal representations by error propagation\" (PDP Vol. 1, Ch. 8).\n\nProblem:\n Inputs: all 2**N binary patterns of length N (N in 2..8).\n Target: parity of the input — 1 if an ...	{ "slug": "n-bit-parity", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/n-bit-parity/n_bit_parity.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nXOR backprop demo, faithful to Rumelhart, Hinton & Williams (1986),\n\"Learning representations by back-propagating errors\", Nature 323, 533-536\n(short version of PDP Vol. 1, Ch. 8).\n\nProblem:\n 4 input/output patterns:\n (0, 0) -> 0\n (0, 1) -> 1\n ...	{ "slug": "xor", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/xor/xor.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nBinary addition of two 2-bit numbers (Rumelhart, Hinton & Williams 1986).\n\nSource:\n Rumelhart, Hinton & Williams (1986),\n \"Learning internal representations by error propagation\",\n PDP Vol. 1, Ch. 8.\n\nProblem:\n Inputs: 4 bits = (a1, a0, b1, b0) repr...	{ "slug": "binary-addition", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/binary-addition/binary_addition.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nSunspots time-series prediction with soft weight-sharing\n(Nowlan & Hinton 1992, \"Simplifying Neural Networks by Soft Weight-Sharing\",\nNeural Computation 4(4), 473-493).\n\nData: yearly Wolfer / SILSO sunspot count, 1700 onwards. Weigend benchmark\nsplit: train 1700-1...	{ "slug": "sunspots", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/sunspots/sunspots.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nDipole-position population code — reproduction of Zemel & Hinton,\n\"Learning population codes by minimizing description length\",\nNeural Computation 7 (1995).\n\nProblem\n-------\nRender 8x8 images of \"dipoles\": a +1 pixel at column x, row y, and a -1 pixel\nat colum...	{ "slug": "dipole-position", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/dipole-position/dipole_position.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nForward-Forward on CIFAR-10 with locally-connected layers.\n\nReproduction of Hinton (2022), \"The Forward-Forward Algorithm: Some Preliminary\nInvestigations\", section on CIFAR-10 — the experiment that argues FF closes\nthe gap with backprop on cluttered images using *...	{ "slug": "ff-cifar-locally-connected", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/ff-cifar-locally-connected/ff_cifar_locally_connected.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nRandom-dot stereograms with the Imax mutual-information objective\n(Becker & Hinton 1992, Nature).\n\nSetup\n A 1-D \"world\" of dots is rendered into two retinal arrays (left and right\n eye). The right eye's view is the same dot pattern shifted horizontally by\n a p...	{ "slug": "random-dot-stereograms", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/random-dot-stereograms/random_dot_stereograms.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nUnsupervised flow capsules on the Geo dataset (synthetic 2D shapes moving\nwith known per-shape affines).\n\nReference: Sabour, Tagliasacchi, Yazdani, Hinton & Fleet,\n\"Unsupervised part representation by flow capsules\", ICML 2021.\n\nArchitecture (numpy implementation...	{ "slug": "geo-flow-capsules", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/geo-flow-capsules/geo_flow_capsules.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nDipole what / where (Zemel & Hinton 1995).\n\nThe first explicit \"what / where\" toy in Hinton's corpus. We render 8x8 binary\nimages of either a horizontal bar (at random row y) or a vertical bar (at\nrandom column x), 50/50 mix. A 64-100-2-100-64 autoencoder is tr...	{ "slug": "dipole-what-where", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/dipole-what-where/dipole_what_where.py" }
hinton_problems	[ { "role": "user", "content": "Implement the following learning problem from scratch in pure NumPy. Produce a single self-contained Python file containing the dataset, the model, the training loop, and an evaluation that prints the headline metric. The implementation must run on a laptop CPU in under five mi...	code	{ "style": "model", "ground_truth": "\"\"\"\nForward-Forward supervised MNIST with the label encoded in the first 10 pixels.\n\nReproduction of Hinton (2022), \"The Forward-Forward Algorithm: Some Preliminary\nInvestigations\", section 3.3 (\"Using FF to model top-down effects in\nperception\" -- the supervised lab...	{ "slug": "ff-label-in-input", "file": "/Users/anton/source/extropic/_posttraining/data_source/hinton-problems/ff-label-in-input/ff_label_in_input.py" }

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.