Paper, researchers, and primary source
Major lab research / efficient_training
The Lottery Ticket Hypothesis finds sparse subnetworks inside randomly initialized dense networks that can train effectively when reset to their original initialization.
Contribution 1
Defined winning tickets as sparse initialized subnetworks capable of matching dense training.
Contribution 2
Used iterative magnitude pruning to discover trainable subnetworks.
Contribution 3
Demonstrated the phenomenon across fully connected and convolutional networks.
Research context
efficient_training / 2018
The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks places lottery ticket inside the broader efficient training discussion at Massachusetts Institute of Technology, with sparsity supplying a second analytical lens. The editorial sequence connects three claims: Defined winning tickets as sparse initialized subnetworks capable of matching dense training; Used iterative magnitude pruning to discover trainable subnetworks; and Demonstrated the phenomenon across fully connected and convolutional networks. The combination matters because pruning only has meaning under the paper's stated setup. Operationally, the record points to one consequence: trainable sparse structure can exist before training, but discovery cost and behavior at modern foundation-model scale are not settled by the original experiments.
Methods and evidence reading
1 cataloged method notes
Method 1
The experimental design in The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks tests defined winning tickets as sparse initialized subnetworks capable of matching dense training and used iterative magnitude pruning to discover trainable subnetworks against the paper's documented baselines, datasets, model variants, or systems workloads.
How to read the evidence
For The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks, the useful question is what the documented setup supports and where transfer remains untested. Its evidence chain begins with two documented moves: Defined winning tickets as sparse initialized subnetworks capable of matching dense training; and Used iterative magnitude pruning to discover trainable subnetworks. Reported evidence then addresses: Demonstrated the phenomenon across fully connected and convolutional networks. The resulting interpretation is practical but conditional: trainable sparse structure can exist before training, but discovery cost and behavior at modern foundation-model scale are not settled by the original experiments. Its boundary is that evidence for lottery ticket in The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks covers architecture choices, evaluation protocol, comparison baselines, compute budget, task distribution, and documented data; behavior beyond that documented envelope remains untested. Any extension should report how altered sparsity conditions affect the original lottery ticket result. An independent check of The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks needs a fixed lottery ticket comparison, a declared sparsity variation, and saved cases where Used iterative magnitude pruning to discover trainable subnetworks does not carry over.
Findings in the source record
1 paper-specific findings
- The reported evidence in The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks supports demonstrated the phenomenon across fully connected and convolutional networks.
Practical implication for AI builders
Massachusetts Institute of Technology / 2018
Proposed BrokenGPT application
Research blueprint / proposed status
Proposed route placement / lottery ticket
For a proposed BrokenGPT experiment based on The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks, evaluate structured pruning only as an offline compression study, preserving initialization, optimizer, quality curves, and hardware-realized speed rather than sparsity alone. Keep the lottery ticket path isolated, versioned, and attributable to this research record.
Proposed acceptance test / sparsity
Validate the proposed lottery ticket route against the paper's reported outcome: Demonstrated the phenomenon across fully connected and convolutional networks. A proposed The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks gate needs training stability, sparsity, convergence, and transfer quality; its sparsity cases should remain disaggregated from the overall lottery ticket score.
Proposed decision boundary / pruning
Balance compute savings, search effort, and reproducibility before promoting the proposed pruning design. Because reusing the mechanism calls for separate evidence about changed operating conditions, a later model revision, new hardware, another user population, and a different product, not an inference from the original benchmark alone, adoption remains conditional on replay under BrokenGPT's selected model, runtime, and policy configuration.
Limitations, verification, and source
Boundaries recorded with the paper
Limitations
- Evidence for lottery ticket in The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks covers architecture choices, evaluation protocol, comparison baselines, compute budget, task distribution, and documented data; behavior beyond that documented envelope remains untested.
- Reusing the mechanism calls for separate evidence about changed operating conditions, a later model revision, new hardware, another user population, and a different product, not an inference from the original benchmark alone.
PRIMARY SOURCES
- 01The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks
Massachusetts Institute of Technology — Primary primary arXiv paper / 9 March 2018 / Jonathan Frankle, Michael Carbin