Matches in SemOpenAlex for { <https://semopenalex.org/work/W3208082192> ?p ?o ?g. }
Showing items 1 to 81 of
81
with 100 items per page.
- W3208082192 abstract "Hamiltonian Monte Carlo (HMC) is a premier Markov Chain Monte Carlo (MCMC) algorithm for continuous target distributions. Its full potential can only be unleashed when its problem-dependent hyperparameters are tuned well. The adaptation of one such hyperparameter, trajectory length ($tau$), has been closely examined by many research programs with the No-U-Turn Sampler (NUTS) coming out as the preferred method in 2011. A decade later, the evolving hardware profile has lead to the proliferation of personal and cloud based SIMD hardware in the form of Graphics and Tensor Processing Units (GPUs, TPUs) which are hostile to certain algorithmic details of NUTS. This has opened up a hole in the MCMC toolkit for an algorithm that can learn $tau$ while maintaining good hardware utilization. In this work we build on recent advances along this direction and introduce SNAPER-HMC, a SIMD-accelerator-friendly adaptive-MCMC scheme for learning $tau$. The algorithm maximizes an upper bound on per-gradient effective sample size along an estimated principal component. We empirically show that SNAPER-HMC is stable when combined with mass-matrix adaptation, and is tolerant of certain pathological target distribution covariance spectra while providing excellent long and short run sampling efficiency. We provide a complete implementation for continuous multi-chain adaptive HMC combining trajectory learning with standard step-size and mass-matrix adaptation in one turnkey inference package." @default.
- W3208082192 created "2021-11-08" @default.
- W3208082192 creator A5008991598 @default.
- W3208082192 creator A5052225069 @default.
- W3208082192 date "2021-10-22" @default.
- W3208082192 modified "2023-09-27" @default.
- W3208082192 title "Focusing on Difficult Directions for Learning HMC Trajectory Lengths" @default.
- W3208082192 cites W1579788182 @default.
- W3208082192 cites W1987034518 @default.
- W3208082192 cites W2048971218 @default.
- W3208082192 cites W2056760934 @default.
- W3208082192 cites W2059448777 @default.
- W3208082192 cites W2100397110 @default.
- W3208082192 cites W2135973421 @default.
- W3208082192 cites W2138309709 @default.
- W3208082192 cites W2148534890 @default.
- W3208082192 cites W2160065175 @default.
- W3208082192 cites W2432567885 @default.
- W3208082192 cites W2478027467 @default.
- W3208082192 cites W2577537660 @default.
- W3208082192 cites W2896065471 @default.
- W3208082192 cites W2906089687 @default.
- W3208082192 cites W2951059525 @default.
- W3208082192 cites W2963977107 @default.
- W3208082192 cites W2964121744 @default.
- W3208082192 cites W2981400429 @default.
- W3208082192 cites W3157648886 @default.
- W3208082192 hasPublicationYear "2021" @default.
- W3208082192 type Work @default.
- W3208082192 sameAs 3208082192 @default.
- W3208082192 citedByCount "0" @default.
- W3208082192 crossrefType "posted-content" @default.
- W3208082192 hasAuthorship W3208082192A5008991598 @default.
- W3208082192 hasAuthorship W3208082192A5052225069 @default.
- W3208082192 hasConcept C105795698 @default.
- W3208082192 hasConcept C107673813 @default.
- W3208082192 hasConcept C111350023 @default.
- W3208082192 hasConcept C11413529 @default.
- W3208082192 hasConcept C150552126 @default.
- W3208082192 hasConcept C154945302 @default.
- W3208082192 hasConcept C173608175 @default.
- W3208082192 hasConcept C19499675 @default.
- W3208082192 hasConcept C33923547 @default.
- W3208082192 hasConcept C41008148 @default.
- W3208082192 hasConceptScore W3208082192C105795698 @default.
- W3208082192 hasConceptScore W3208082192C107673813 @default.
- W3208082192 hasConceptScore W3208082192C111350023 @default.
- W3208082192 hasConceptScore W3208082192C11413529 @default.
- W3208082192 hasConceptScore W3208082192C150552126 @default.
- W3208082192 hasConceptScore W3208082192C154945302 @default.
- W3208082192 hasConceptScore W3208082192C173608175 @default.
- W3208082192 hasConceptScore W3208082192C19499675 @default.
- W3208082192 hasConceptScore W3208082192C33923547 @default.
- W3208082192 hasConceptScore W3208082192C41008148 @default.
- W3208082192 hasLocation W32080821921 @default.
- W3208082192 hasOpenAccess W3208082192 @default.
- W3208082192 hasPrimaryLocation W32080821921 @default.
- W3208082192 hasRelatedWork W121871126 @default.
- W3208082192 hasRelatedWork W1977702261 @default.
- W3208082192 hasRelatedWork W2086913350 @default.
- W3208082192 hasRelatedWork W2521822742 @default.
- W3208082192 hasRelatedWork W2737551591 @default.
- W3208082192 hasRelatedWork W2750364761 @default.
- W3208082192 hasRelatedWork W2783596265 @default.
- W3208082192 hasRelatedWork W2803936899 @default.
- W3208082192 hasRelatedWork W2811063752 @default.
- W3208082192 hasRelatedWork W2888088597 @default.
- W3208082192 hasRelatedWork W2903361505 @default.
- W3208082192 hasRelatedWork W2905549570 @default.
- W3208082192 hasRelatedWork W2951362160 @default.
- W3208082192 hasRelatedWork W2953125965 @default.
- W3208082192 hasRelatedWork W3003471248 @default.
- W3208082192 hasRelatedWork W3044469112 @default.
- W3208082192 hasRelatedWork W3171305131 @default.
- W3208082192 hasRelatedWork W3197643534 @default.
- W3208082192 hasRelatedWork W3197747679 @default.
- W3208082192 hasRelatedWork W3401598 @default.
- W3208082192 isParatext "false" @default.
- W3208082192 isRetracted "false" @default.
- W3208082192 magId "3208082192" @default.
- W3208082192 workType "article" @default.