Institution profile

MiniMax

Research institutionasia · cn
Official website
Research library8linked papers
Opportunities0open roles
Selected work

Representative Papers

Slow Beats Fast at the Kesten-Stigum Threshold: Minimax, Fisher-Information and Belief-Propagation Characterizations of the Information-Computation Gap in Sparse Stochastic Block Models

Oct 06, 2026

This study addresses the information-computation gap in community recovery within sparse stochastic block models, where the properties of the Kesten-Stigum threshold remain insufficiently characterized. By integrating statistical decision theory with Fisher information, this work proposes a triple characterization framework encompassing minimax risk equivalence, cycle-counting Fisher information convergence, and EM step-size invariance. Theoretical derivations are conducted using Bayesian risk, low-degree polynomial methods, and belief propagation algorithms. This research rigorously establishes the existence of the threshold for two communities (q=2) and identifies the hard phase window for five communities (q=5), elucidating generalization patterns for multi-community extensions. Furthermore, experiments on networks comprising up to 300,000 nodes validate the accuracy of these theoretical predictions.

0 citationsRead paper

Sol-H3: Recursive Self-Improvement for MiniMax-H3 Inference Acceleration on Sol-Engine across Cloud and Edge

Sep 28, 2026

This study addresses the high deployment latency and memory constraints of the MiniMax-H3 video diffusion model, which stem from its massive parameter count and multi-step denoising process, by constructing a full-stack inference acceleration pipeline. Methodologically, it proposes a cross-resolution two-stage scheduling strategy coupled with latent-space mapping to eliminate redundant VAE encoding and decoding operations. Furthermore, a recursive self-improvement (RSI) mechanism is introduced to automatically search for optimal kernel fusion and memory layout configurations. Experimental results demonstrate that this approach achieves up to 30× end-to-end speedup and a 20% reduction in GPU memory consumption. Notably, a single DGX Spark card can complete video generation within one minute, effectively overcoming the computational bottlenecks associated with cloud-based deployment.

0 citationsRead paper

Sharp Critical Minimax Laws and No-Learning Thresholds in Continuous-Time Adaptive Control

Sep 27, 2026

This study addresses minimax-improved adaptation laws and learning thresholds under unknown gains in continuous-time adaptive control. Methodologically, it establishes a uniform deficit-energy inequality within a scalar Gaussian experiment framework, proposes a moving soft-thresholding feedback mechanism, and conducts the analysis through Gaussian sequential control theory. The primary contributions include the first rigorous proof of an exact quartic-logarithmic critical law, revealing phase-transition boundaries for learning and the impact of terminal cost variance. Furthermore, the work establishes the asymptotic task boundary c_gH^4 = d^2/2 and verifies that the regret bound approaches the optimal O(N^{-1/2}) rate for fixed-horizon Gaussian problems.

0 citationsRead paper
Recent publications

Latest Papers

Slow Beats Fast at the Kesten-Stigum Threshold: Minimax, Fisher-Information and Belief-Propagation Characterizations of the Information-Computation Gap in Sparse Stochastic Block Models

Oct 06, 2026

This study addresses the information-computation gap in community recovery within sparse stochastic block models, where the properties of the Kesten-Stigum threshold remain insufficiently characterized. By integrating statistical decision theory with Fisher information, this work proposes a triple characterization framework encompassing minimax risk equivalence, cycle-counting Fisher information convergence, and EM step-size invariance. Theoretical derivations are conducted using Bayesian risk, low-degree polynomial methods, and belief propagation algorithms. This research rigorously establishes the existence of the threshold for two communities (q=2) and identifies the hard phase window for five communities (q=5), elucidating generalization patterns for multi-community extensions. Furthermore, experiments on networks comprising up to 300,000 nodes validate the accuracy of these theoretical predictions.

0 citationsRead paper

Sol-H3: Recursive Self-Improvement for MiniMax-H3 Inference Acceleration on Sol-Engine across Cloud and Edge

Sep 28, 2026

This study addresses the high deployment latency and memory constraints of the MiniMax-H3 video diffusion model, which stem from its massive parameter count and multi-step denoising process, by constructing a full-stack inference acceleration pipeline. Methodologically, it proposes a cross-resolution two-stage scheduling strategy coupled with latent-space mapping to eliminate redundant VAE encoding and decoding operations. Furthermore, a recursive self-improvement (RSI) mechanism is introduced to automatically search for optimal kernel fusion and memory layout configurations. Experimental results demonstrate that this approach achieves up to 30× end-to-end speedup and a 20% reduction in GPU memory consumption. Notably, a single DGX Spark card can complete video generation within one minute, effectively overcoming the computational bottlenecks associated with cloud-based deployment.

0 citationsRead paper

Sharp Critical Minimax Laws and No-Learning Thresholds in Continuous-Time Adaptive Control

Sep 27, 2026

This study addresses minimax-improved adaptation laws and learning thresholds under unknown gains in continuous-time adaptive control. Methodologically, it establishes a uniform deficit-energy inequality within a scalar Gaussian experiment framework, proposes a moving soft-thresholding feedback mechanism, and conducts the analysis through Gaussian sequential control theory. The primary contributions include the first rigorous proof of an exact quartic-logarithmic critical law, revealing phase-transition boundaries for learning and the impact of terminal cost variance. Furthermore, the work establishes the asymptotic task boundary c_gH^4 = d^2/2 and verifies that the regret bound approaches the optimal O(N^{-1/2}) rate for fixed-horizon Gaussian problems.

0 citationsRead paper