?url_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Adc&rft.title=Thermostat-assisted+continuously-tempered+Hamiltonian+Monte+Carlo+for+Bayesian+learning&rft.creator=Luo%2C+R&rft.creator=Wang%2C+J&rft.creator=Yang%2C+Y&rft.creator=Zhu%2C+Z&rft.creator=Wang%2C+J&rft.description=We+propose+a+new+sampling+method%2C+the+thermostat-assisted+continuously-tempered%0D%0AHamiltonian+Monte+Carlo%2C+for+Bayesian+learning+on+large+datasets+and+multimodal%0D%0Adistributions.+It+simulates+the+Nos%C3%A9-Hoover+dynamics+of+a+continuously-tempered%0D%0AHamiltonian+system+built+on+the+distribution+of+interest.+A+significant+advantage+of%0D%0Athis+method+is+that+it+is+not+only+able+to+efficiently+draw+representative+i.i.d.+samples%0D%0Awhen+the+distribution+contains+multiple+isolated+modes%2C+but+capable+of+adaptively%0D%0Aneutralising+the+noise+arising+from+mini-batches+and+maintaining+accurate+sampling.%0D%0AWhile+the+properties+of+this+method+have+been+studied+using+synthetic+distributions%2C%0D%0Aexperiments+on+three+real+datasets+also+demonstrated+the+gain+of+performance+over%0D%0Aseveral+strong+baselines+with+various+types+of+neural+networks+plunged+in.&rft.publisher=Neural+Information+Processing+Systems+Foundation%2C+Inc.&rft.contributor=Bengio%2C+S&rft.contributor=Wallach%2C+H&rft.contributor=Larochelle%2C+H&rft.contributor=Grauman%2C+K&rft.contributor=CesaBianchi%2C+N&rft.contributor=Garnett%2C+R&rft.date=2018-12-08&rft.type=Proceedings+paper&rft.publisher=32nd+Conference+on+Neural+Information+Processing+Systems+(NIPS)&rft.language=eng&rft.source=+++++In%3A+Bengio%2C+S+and+Wallach%2C+H+and+Larochelle%2C+H+and+Grauman%2C+K+and+CesaBianchi%2C+N+and+Garnett%2C+R%2C+(eds.)+Advances+In+Neural+Information+Processing+Systems+31+(Nips+2018).++++Neural+Information+Processing+Systems+Foundation%2C+Inc.%3A+Montreal%2C+Canada.+(2018)+++++&rft.format=text&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F10079536%2F1%2F8266-thermostat-assisted-continuously-tempered-hamiltonian-monte-carlo-for-bayesian-learning.pdf&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F10079536%2F&rft.rights=open