Search
Search Funnelback University
- Refined by:
- Date: 2018
31 -
80 of
555
search results for news |u:mi.eng.cam.ac.uk
Fully-matching results
-
Uncertainty management for on-line optimisation of a…
mi.eng.cam.ac.uk/~sjy/papers/dgcg11.pdf20 Feb 2018: 1 and a (uniform) ran-dom action otherwise. The obtained transition is then used tolearn the new estimate Q̂i1 (and σ̂i1). ... The parameters used in the two new explorationschemes were set as follows: α = 12, β = 1000 and β0 = 100.These were -
BAYESIAN UPDATE OF DIALOGUE STATE FOR ROBUST DIALOGUE SYSTEMS ...
mi.eng.cam.ac.uk/~sjy/papers/thsy08.pdf20 Feb 2018: This allows asimple trade off between the emphasis placed on new informationcompared to old accumulated beliefs. ... This paper has introduced a new Bayesian Update of Dialogue Stateapproach to maintaining beliefs in dialogue systems. -
A New Look at Filtering Techniques for Illumination Invariance ...
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/2006-AFGR-Arandjelovic-filtering.pdf13 Mar 2018: A New Look at Filtering Techniques for Illumination Invariance in AutomaticFace Recognition. -
./plot_entropy.eps
mi.eng.cam.ac.uk/~ar527/chen_is2017.pdf15 Jun 2018: The performance of the bi-RNNLMs is evaluatedon three speech recognition tasks: broadcast news; meetingtran-scription (AMI); and low-resource systems (Babel data). ... This projected, word vectorxt1 is then combined with thehistory vectorht2, which -
User study of the Bayesian Update of Dialogue State ...
mi.eng.cam.ac.uk/~sjy/papers/tgkm0820 Feb 2018: In previous experiments the user simulator would change its goaland require a venue matching the new goal but this is difficult to test ina real environment. -
yeyo05b.dvi
mi.eng.cam.ac.uk/~sjy/papers/yeyo05b.pdf20 Feb 2018: The results show thatusing a pool of non-native adaptation data collected in advance,can substantially improve the performance on new non-nativespeakers. ... Speaker depen-dent adaptation might be expected to provide further improve-ment, however as -
ROBUST DIALOG STATE TRACKING USING DELEXICALISEDRECURRENT NEURAL…
mi.eng.cam.ac.uk/~sjy/papers/htyo14.pdf20 Feb 2018: New slots Old slots Joint. meanUnadapted 0.866 0.877 0.552. Adapted 0.869 0.890 0.566. ... Ac-curacy is reported on the Old slots and New slots, i.e. -
yokou
mi.eng.cam.ac.uk/~sjy/papers/toyo09.pdf20 Feb 2018: Inadifferentapproach to improving thesyntheticspeechquality,Toda and Tokuda [5] have introduced a new criterion on a higher-order moment called the global variance (GV), which is the vari-ance of the static ... 6. CONCLUSIONSThis paper has described a -
Reward-Balancing for Statistical Spoken Dialogue Systems…
mi.eng.cam.ac.uk/~sjy/papers/ubcy17.pdf20 Feb 2018: To train a policy using multi-objective GP-SARSA, a new weight configuration is sampledrandomly for each training dialogue. -
Enginuity
mi.eng.cam.ac.uk/~cipolla/archive/Publications/inproceedings/1997-Enginuity-face-detection.pdf13 Mar 2018: A new approach. An engineering viewpoint. This is part of a more ambitious project on understanding vision and making machines see. -
Active Memory Networks for Language Modeling O. Chen, A. ...
mi.eng.cam.ac.uk/~ar527/chen_is2018.pdf15 Jun 2018: Experimentswere conducted on the Penn Tree Bank and BBC Multi-GenreBroadcast News (MGB) corpora, where the proposed approachsignificantly outperforms standard forms of recurrent models inperplexity. ... Experiments conducted on PTB and the MGBtasks show -
paper.dvi
mi.eng.cam.ac.uk/~sjy/papers/yswy07.pdf20 Feb 2018: Based on the current belief stateb, the machine selects anaction am Am, receives a rewardr(sm, am), and transitions toa new (unobserved) states′m, wheres. ′. ... mass of 1.0 and this is redistributed according to the prior in thecorresponding -
MVA2000 lAPR Workshop on Machine Vision Applications. Nov.…
mi.eng.cam.ac.uk/~cipolla/archive/Publications/inproceedings/2000-MVA-Inoue.pdf13 Mar 2018: Then, the tracking is operated by the following sequence. 1. Capture the new video h a m e and render. ... feature points to the image plane. 2. Search for new positions of the feature. -
WORD-LEVEL EMPHASIS MODELLING IN HMM-BASED SPEECH SYNTHESIS K. Yu, ...
mi.eng.cam.ac.uk/~sjy/papers/yumy10.pdf20 Feb 2018: Two new decision tree cluster-ing approaches, two-pass and factorized decision tree, are proposedto effectively address this problem. ... In thispaper, two new approaches are investigated to address the problem.In the first method, the state-clustering -
MVAZOOO lAPR Workshop on Machine Vision Applications. Nov.…
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/2000-MVA-Infantino.pdf13 Mar 2018: viewpoint. Finally, triangulation is used t o recover 3D models of the scene and to visu- alise new viewpoints. ... Decomposition of inter-frame homography matrix gives the new es- timation of the projection matrices of two views. -
Transformation-based Learning for Semantic parsing F. Jurčı́ček, M. …
mi.eng.cam.ac.uk/~sjy/papers/jgkm09.pdf20 Feb 2018: New n-gram features were generated in which aword history is given by links between words. ... pairs thatappear in the output semantics (e.g., from.city = New York). -
A Clustering Approach to Semantic Decoding Hui Ye and ...
mi.eng.cam.ac.uk/~sjy/papers/yeyo06b20 Feb 2018: Thus, more than one template can con-tribute to the set of slot/value pairs extracted from an utterance.This allows new forms of sentence to be processed by partiallymatching against ... Furthermore,for real applications, Y-clustering is extremely -
paper.dvi
mi.eng.cam.ac.uk/~ar527/ragni_is2018a.pdf15 Jun 2018: 1. IntroductionSpeech data present in the “wild“ comes in various forms suchas Youtube, podcasts and radio news. ... Xiang, S. Abdou, and J. Makhoul, “Re-cent progress in Arabic broadcast news transcription at BBN,” inInterspeech, 2005. -
A DATA-DRIVEN SPOKEN LANGUAGE UNDERSTANDING SYSTEM Yulan He Steve ...
mi.eng.cam.ac.uk/~sjy/papers/heyo03b.pdf20 Feb 2018: by positions followed by a push of one ormore new preterminal semantic concepts relating to the next in-put word. ... of the 32nd Annual Meeting of the Association forComputational Linguistics, Las Cruces, New Maxico, June1994, pp. -
A Framework for Dialogue Data Collection with a Simulated ...
mi.eng.cam.ac.uk/~sjy/papers/stwy04.pdf20 Feb 2018: Ideally therewill be little overhead associated with undertaking a new col-lection. -
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/2005-MVA-Conde.pd…
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/2005-MVA-Conde.pdf13 Mar 2018: This database is not yet completed as more data are being acquired for new subjects, and, in order to evaluate changes in the facial expression through time, new sets of images ... For instance, all noses have a similar shape, consequently it is not -
robust.dvi
mi.eng.cam.ac.uk/~sjy/papers/heyo04.pdf20 Feb 2018: In order, to limit the issues toparameter mismatch problems, the new application cho-sen (Communicator) covered essentially the same set of. ... that they are robust and they are read-ily adaptable to new or changing applications. -
Towards Using Conversations with Spoken Dialogue Systems in…
mi.eng.cam.ac.uk/~sjy/papers/lygk16.pdf20 Feb 2018: mapped to a new intermediatelevel placed between the original levels in the or-dered scale. -
Photo-Realistic Expressive Text to Talking Head Synthesis Vincent…
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/2013-Interspeech-Talking-Head.pdf13 Mar 2018: Subtle, extreme and new combined ex-pressions are achieved by moving around the space. ... 14] Sky News, “Red Dwarf talking head could be a future PA”, 19March 2013. -
MVA '96 IAPR Workshop on Machine Vision Applications. November.…
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/1996-MVA-affine-servo.pdf13 Mar 2018: The technique is based on a circular search region around the image centroid, along which a new ill- tensity profile is established. -
ON-LINE POLICY OPTIMISATION OF BAYESIAN SPOKEN DIALOGUE SYSTEMS…
mi.eng.cam.ac.uk/~sjy/papers/gbhk13.pdf20 Feb 2018: 7: if non-terminal step then8: Choose new action a′ π(c′) (Eq. ... This new optimisation frameworkavoids the need for building costly application-dependent user sim-ulators and yields improved run-time performance, opening the wayfor building -
Back-off Action Selection in Summary Space-BasedPOMDP Dialogue…
mi.eng.cam.ac.uk/~sjy/papers/gljk09.pdf20 Feb 2018: Every time a new beliefstate is visited, it is mapped to a summary state and then tothe nearest grid point. ... j) 1. 20: else create new pt21: add b̂t to B22: Q(b̂t, ait. -
Real user evaluation of spoken dialogue systems using Amazon ...
mi.eng.cam.ac.uk/~sjy/papers/jkgm11.pdf20 Feb 2018: 1. IntroductionDespite recent progress in spoken dialogue systems develop-ment, there is a re-occurring problem with evaluating new ideasin this area. ... If auser changes his mind during the dialogue, this is accounted forby identifying a new goal that -
SynthCam3D: Semantic Understanding With Synthetic Indoor Scenes Ankur …
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/2015-CVPR-SynthCam3D.pdf13 Mar 2018: We make publicly available a new library – Synth-Cam3D – consisting of a significant number of labelled syn-thetic 3D scenes and associated code for generating depthmaps and their corresponding annotations. ... Future work includes expandingthe -
Multi-domain Neural Network Language Generation forSpoken Dialogue…
mi.eng.cam.ac.uk/~sjy/papers/wgmr16.pdf20 Feb 2018: Repeat this process for every slotin the instance and yield a new pseudo instance(d̂i, ̂i) T in the target domain. ... Goodfellow, Arnaud Berg-eron, Nicolas Bouchard, and Yoshua Bengio. 2012.Theano: new features and speed improvements. -
Training a real-world POMDP-based Dialogue System Blaise Thomson,…
mi.eng.cam.ac.uk/~sjy/papers/tswy07.pdf20 Feb 2018: Online training allows the system to adaptto real users as new dialogues are recorded. ... After a batch of dialogues have beencompleted, the estimates of the Q-functions are updatedwith the new dialogue scores. -
tech.dvi
mi.eng.cam.ac.uk/~sjy/papers/bghk13.pdf20 Feb 2018: likelihood. Otherclassifiers such as SVMs [13] and MLPs [14] have also been used.This classification approach has been used successfully in LVCSRtasks such as meetings [15, 14, 16] and broadcast news ... 17] J.L. Gauvain, L. Lamel, and G. Adda, -
Paper Preparation and Submission Instructions for EUSIPCO 2004
mi.eng.cam.ac.uk/~sjy/papers/poyo06.pdf20 Feb 2018: It is today well known that after laryngectomy the pharyngoesophageal segment acts as the new voice source, i.e. -
Uncertain RanSaC Ben Tordoff and Roberto CipollaDepartment of…
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/2005-MVA-Tordoff.pdf13 Mar 2018: Section 4 compares the per-formance of this new scheme with the standard model be-fore conclusions are drawn. ... 13] P. H. S. Torr and A. Zisserman. MLESAC: A new robust esti-mator with application to estimating image geometry. -
Learning from Real Users: Rating Dialogue Success with Neural ...
mi.eng.cam.ac.uk/~sjy/papers/svgk15.pdf20 Feb 2018: Currently work is focused on investigating less domain spe-cific features, the dependence on the simulated user, transfer-ring the RNN models to new domains, and using them for re-ward ... Bergstra, I. J. Goodfellow,A. Bergeron, N. Bouchard, and Y. Bengio -
THE USE OF DISCRIMINATIVE BELIEF TRACKINGIN POMDP-BASED DIALOGUE…
mi.eng.cam.ac.uk/~sjy/papers/khgt14.pdf20 Feb 2018: Therefore in one turn b and m areupdated to give the new belief b′ and memory m′. ... The above networks applied to tagged and untagged in-puts are combined to give the new belief b′:. -
Scaling Up POMDPs for Dialog Management: The “Summary POMDP” ...
mi.eng.cam.ac.uk/~sjy/papers/wiyo05.pdf20 Feb 2018: This would result in a significant reduction in )(1 londonbt , and a new most. ... 16] Zhang Bo, Cai Qingsheng, Mao Jianfeng, and Guo Baining. Planning and Acting under Uncertainty: A New Model for Spoken Dialogue System. -
A Statistical Consistency Check for the SpaceCarving Algorithm. A. ...
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/2000-BMVC-Broadhurst-consistency.pdf13 Mar 2018: A new consistencyfunction is proposedthat usesa statisticalcomparisoninsteadof the voxelcentroidsamplingthat wasinitially proposed.This is importantwhenthereis moredetailin theimagesthancanbestoredin a voxel representation.Thenew functionis -
POLICY COMMITTEE FOR ADAPTATION IN MULTI-DOMAIN SPOKEN…
mi.eng.cam.ac.uk/~sjy/papers/gmsv15.pdf20 Feb 2018: 3. a flexible architecture which allows for both adding andremoving committee members, which particularly use-ful for extending multi-domain systems to new unseendomains. ... Since the Gaussian processfor the Q-function defines a Gaussian distribution -
MVAZOOO lAPR Workshop on Machine Vision Applications. Nov.…
mi.eng.cam.ac.uk/~cipolla/archive/Publications/inproceedings/2000-MVA-Infantino.pdf13 Mar 2018: viewpoint. Finally, triangulation is used t o recover 3D models of the scene and to visu- alise new viewpoints. ... Decomposition of inter-frame homography matrix gives the new es- timation of the projection matrices of two views. -
MVA'94 IAPR Workshop on Machine Vision Applications Dec. 13-15,…
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/1994-MVA-pointing.pdf13 Mar 2018: 0, O), (0, I ) , ( I , I ) , ( 1 , O)), we define a new coordinate system on the plane, which we call working plane coordinates. -
N-BEST ERROR SIMULATION FOR TRAINING SPOKEN DIALOGUE SYSTEMS Blaise…
mi.eng.cam.ac.uk/~sjy/papers/thgt12.pdf20 Feb 2018: Thehope is that these new error models will be able to improve the re-sulting performance of trained dialogue systems. ... When using the new error simulatorsto train a system, one would expect that the resulting system shouldthen perform better with -
MULTI-DOMAIN DIALOGUE SUCCESS CLASSIFIERS FOR POLICY TRAINING David…
mi.eng.cam.ac.uk/~sjy/papers/vsgm15.pdf20 Feb 2018: Pooling and adaptation arealso investigated for constructing multi-domain models whendata is available in the new domain. ... success classifiers in order to rapidly obtain suitableclassification performance in this new domain. -
POMDP-based dialogue manager adaptation to extended domains M.…
mi.eng.cam.ac.uk/~sjy/papers/gbhk13a.pdf20 Feb 2018: 4 will be reduced and thus leadto the subsequent optimisation in the new targetdomain falling prematurely into a local optimum. ... 3). 8 Conclusions. This paper has investigated the problem of ex-tending a dialogue system to handle new previ-ously -
SynthCam3D: Semantic Understanding With Synthetic Indoor Scenes Ankur …
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/2015-CVPR-SynthCam3D.pdf.pdf13 Mar 2018: We make publicly available a new library – Synth-Cam3D – consisting of a significant number of labelled syn-thetic 3D scenes and associated code for generating depthmaps and their corresponding annotations. ... Future work includes expandingthe -
RECONSTRUCTION OF SCULPTURE FROM UNCALIBRATED IMAGE PROFILES Roberto…
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/2001-ICIP-Cipolla-sculpture.pdf13 Mar 2018: The circular motion will generate a web of contour gen-erators around the object, which can be exploited for reg-istering any new arbitrary general view. ... 2). j0. e j1 e j2. e. Fig. 2. Three views from circular motion provide 6 outerepipolar tangents -
Training and Evaluation of the HIS POMDP Dialogue System ...
mi.eng.cam.ac.uk/~sjy/papers/gkms08.pdf20 Feb 2018: Based on this current beliefstate b, the machine selects an action am, receivesa reward r(sm, am), and transitions to a new (un-observed) state s′m, where s′m depends ... If the nextsystem action is to make an alternative offer, thenthe new -
Tracking Using Online Feature Selectionand a Local Generative Model…
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/2007-BMVC-Woodley.pdf13 Mar 2018: The target estimate is used as a new positive example and surrounding regions asnegative examples. ... Row 2 shows corresponding results with the new algorithm. Regionsdetected as occluded are coloured white. -
Stochastic Language Generation in Dialogue using Recurrent…
mi.eng.cam.ac.uk/~sjy/papers/wgkm15.pdf20 Feb 2018: trees. Objective metricssuggest that this new model outperformsprevious methods under the same experi-mental conditions. ... Goodfellow, Arnaud Berg-eron, Nicolas Bouchard, and Yoshua Bengio. 2012.Theano: new features and speed improvements.Deep Learning -
1620IEICE TRANS. INF. & SYST., VOL.E84–D, NO.12 DECEMBER 2001 ...
mi.eng.cam.ac.uk/~cipolla/archive/Publications/article/2001-IAS-Infantino.pdf13 Mar 2018: Finally, triangulation is used to recover 3D mod-els of the scene and to visualise new viewpoints. ... Decomposition of inter-frame homography ma-trix gives the new estimation of the projection matricesof two views (Fig.
Search history
Recently clicked results
Recently clicked results
Your click history is empty.
Recent searches
Recent searches
Your search history is empty.