Jonathan Frankle

Beliefs Korrents

• Korrents What they believe 5 beliefs — each backed by an exact quote.

  Each is a korrent — compiled by korrents.com, not by them: the one-line wordings are korrents', the quotes are theirs.

  Recent

  • Pruning cuts the parameters of a trained network by over 90% without costing accuracy, so inference gets smaller and faster for free.

    Neural network pruning techniques can reduce the parameter counts of trained networks by over 90%, decreasing storage requirements and improving computational performance of inference without compromising accuracy.

    Frankle & Carbin, "The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks" (arXiv) Said 9 Mar 2018

    Full record — everyone who holds this →

    neural networkspruning

  • The saving pruning offers is only ever at inference, because the sparse architectures it produces are hard to train from the start.

    However, contemporary experience is that the sparse architectures produced by pruning are difficult to train from the start, which would similarly improve training performance.

    Frankle & Carbin, "The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks" (arXiv) Said 9 Mar 2018

    Full record — everyone who holds this →

    pruning

  • Pruning does not merely shrink a network; it uncovers subnetworks whose original initialization was what made them trainable at all.

    We find that a standard pruning technique naturally uncovers subnetworks whose initializations made them capable of training effectively.

    Frankle & Carbin, "The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks" (arXiv) Said 9 Mar 2018

    Full record — everyone who holds this →

    pruning

  Show 2 more

  • A winning ticket wins on its initial weights: the connections it keeps started at values that happen to make training work.

    The winning tickets we find have won the initialization lottery: their connections have initial weights that make training particularly effective.

    Frankle & Carbin, "The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks" (arXiv) Said 9 Mar 2018

    Full record — everyone who holds this →

    pruning

  • A small enough winning ticket learns faster than the network it was cut out of, and ends up more accurate than it.

    Above this size, the winning tickets that we find learn faster than the original network and reach higher test accuracy.

    Frankle & Carbin, "The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks" (arXiv) Said 9 Mar 2018

    Full record — everyone who holds this →

    pruning

  Beliefs others hold too

  • Pruning cuts the parameters of a trained network by over 90% without costing accuracy, so inference gets smaller and faster for free. 2 hold this

    Neural network pruning techniques can reduce the parameter counts of trained networks by over 90%, decreasing storage requirements and improving computational performance of inference without compromising accuracy.

    Frankle & Carbin, "The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks" (arXiv) Said 9 Mar 2018

    Full record — everyone who holds this →

    neural networkspruning

  • The saving pruning offers is only ever at inference, because the sparse architectures it produces are hard to train from the start. 2 hold this

    However, contemporary experience is that the sparse architectures produced by pruning are difficult to train from the start, which would similarly improve training performance.

    Frankle & Carbin, "The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks" (arXiv) Said 9 Mar 2018

    Full record — everyone who holds this →

    pruning

  • Pruning does not merely shrink a network; it uncovers subnetworks whose original initialization was what made them trainable at all. 2 hold this

    We find that a standard pruning technique naturally uncovers subnetworks whose initializations made them capable of training effectively.

    Frankle & Carbin, "The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks" (arXiv) Said 9 Mar 2018

    Full record — everyone who holds this →

    pruning

  All 5 beliefs →