Research map / Learning theory and algorithms / Submodular and game algorithms
Deep Learning Games
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| regret matching (RM) with L1-ball constraint and large random initialization of cumulative regrets used as training method, instead of SGD/Adam/RMSprop | deep feedforward networks trained as deep learning game · MNIST fully connected and convolutional | convergence rate and test misclassification error | Mixed |
| regret matching with L1 constraint | deep networks trained as games · MNIST fully connected and convolutional | fraction of zeroed parameter matrices (sparsity): zeroed 32%, 26%, 63% (FC); 29%, 27%, 28%, 43% (conv) | Helped |
| regret matching without random initialization of cumulative regret vectors (learns too quickly, top layers become sparse) (ablation) | RM for deep learning games · deep learning game experiments | training behavior | Hurt |
| exponentiated weights (EWA) and projected SGD in place of regret matching for deep learning games (ablation) | deep learning game training · deep learning game experiments | training competitiveness | Hurt |
| regret matching vs EWA and projected SGD for one-layer L1-constrained softmax training (ablation) | one-layer learning problem on MNIST and synthetic data · synthetic and MNIST training loss | training loss of averaged protagonist strategy | Helped |
| regret matching in deep model training vs SGD-based methods on fully connected MNIST | RM runtime · Tensorflow implementation | run time relative to RMSProp: 7% slower on convolutional; 85% slower on fully connected | Hurt |
Closest papers by meaning
- Solving Neural Min-Max Games: The Role of Architecture, Initialization & Dynamics (NeurIPS 2025 · similarity 0.89)
- Learning Not to Regret (AAAI 2024 · similarity 0.89)
- No-regret learning in games with noisy feedback: Faster rates and adaptivity via learning rate separation (NeurIPS 2022 · similarity 0.88)
- Near-Optimal No-Regret Learning Dynamics for General Convex Games (NeurIPS 2022 · similarity 0.88)
- On Tractable $\Phi$-Equilibria in Non-Concave Games (NeurIPS 2024 · similarity 0.86)
- A representation-learning game for classes of prediction tasks (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.