Search

29 Apr 2024: References. [1] Control-DAG: Constrained decoding for non-autoregressive directed acyclic T5 using weighted finite state automata, Jinghong Chen, Weizhe Lin, Jingbiao Mei, and Bill Byrne, 2024. To appear at NAACL’24. The Directed Acyclic

5 Jun 2002: MPE Training Results (I). Train eval98 eval98 changeMLE 41.8 46.6 –MMIE 30.1 44.3 -2.3MMIE (τ =200) 32.2 43.8 -2.8MPE (τ =50) 27.9 ... Train eval98 eval98 changeMLE baseline 47.2 45.6 –MMIE 37.7 41.8 -3.8MMIE (τ =200) 35.8 41.4 -4.2MPE (τ =100) 34

10 May 2000: BRF is run on the parallel collectionadding the best 200 new terms taken from the top 10 doc-uments to the original document with a term frequencyequal to 1.

29 Jul 2001: Different values oft and rwere investigated for the document expansion stage, but thereproved to be little difference between the results, so the valuesof t = 200, r = 10 were chosen to be ... 31 51.31s1 - - 12 26 59.11 57.14 54.04 50.65s1 200 10 - -

29 Apr 2011: b) constant density. 0 100 200 300 400 500 600 700 8000.5. ... 0 100 200 300 400 500 600 700 8000.5. 0.625. 0.8.

10 Oct 2001: 100 terms with DBRF1 t=200, r=10, tf=1 ; DBRF2 t=400, r=10, tf=1 ; DBRF t=200, r=20, tf=0.5 Parallel corpus recorded pre test collection, ... t=200 r=20, tf=0.5 (same as T8) X: For Trec-8 -> Nov 96 to Jan 98 (pre-trec-8) = 60,000 stories (129614 terms).

15 Jun 2018: 1. 2. 3. 4. 5. 6. 7. 8. 0 100 200 300 400 500 600.

19 Apr 2017: Thisarchitecture is a Deep Density Network (DDN),. 3. 200. 201. 202.

13 Mar 2018: Figure 2:Parallax used to cluster input face images (a). The distributions ofη (1) for the three clusters, computedfrom 200 manually labelled frames is shown in (b). ... 20. 40. 60. 80. 100. 120. (a) (b). 0 100 200 300 400 500 600 700 800 900 10000.

18 May 2017: 4. 6. 0 50 100 150 200 250 300Instances. 0.3. 0.4. ... 4. 2. 0. 2. 4. 6. 0 50 100 150 200 250 300Instances.