viterbi training improves
play

Viterbi Training Improves Unsupervised Dependency Parsing Valentin - PowerPoint PPT Presentation

Oct 24, 2023 •378 likes •2.17k views

Viterbi Training Improves Unsupervised Dependency Parsing Valentin I. Spitkovsky with Hiyan Alshawi (Google Inc.) Daniel Jurafsky (Stanford University) and Christopher D. Manning (Stanford University) Spitkovsky et al. (Stanford & Google)

  1. The Problem Learning Learning: EM, via inside-outside re-estimation sentences { s } , legal parse trees t ∈ T ( s ) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 7 / 26

  2. The Problem Learning Learning: EM, via inside-outside re-estimation sentences { s } , legal parse trees t ∈ T ( s ) , and a gold t ∗ Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 7 / 26

  3. The Problem Learning Learning: EM, via inside-outside re-estimation sentences { s } , legal parse trees t ∈ T ( s ) , and a gold t ∗ non-convex objective — very sensitive to initialization Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 7 / 26

  4. The Problem Learning Learning: EM, via inside-outside re-estimation sentences { s } , legal parse trees t ∈ T ( s ) , and a gold t ∗ non-convex objective — very sensitive to initialization maximizing the probability of data (sentence strings): � � ˆ θ UNS = arg max P θ ( t ) θ s t ∈ T ( s ) � �� � P θ ( s ) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 7 / 26

  5. The Problem Learning Learning: EM, via inside-outside re-estimation sentences { s } , legal parse trees t ∈ T ( s ) , and a gold t ∗ non-convex objective — very sensitive to initialization maximizing the probability of data (sentence strings): � � ˆ θ UNS = arg max P θ ( t ) θ s t ∈ T ( s ) � �� � P θ ( s ) supervised objective would be convex (counting): � ˆ P θ ( t ∗ ( s )) θ SUP = arg max θ s Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 7 / 26

  6. The Data WSJ Standard Corpus: WSJ k Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 8 / 26

  7. The Data WSJ Standard Corpus: WSJ k The Wall Street Journal section of the Penn Treebank Project (Marcus et al., 1993) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 8 / 26

  8. The Data WSJ Standard Corpus: WSJ k The Wall Street Journal section of the Penn Treebank Project (Marcus et al., 1993) ◮ ... stripped of punctuation, etc. Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 8 / 26

  9. The Data WSJ Standard Corpus: WSJ k The Wall Street Journal section of the Penn Treebank Project (Marcus et al., 1993) ◮ ... stripped of punctuation, etc. ◮ ... rid of sentences left with more than k POS tags; Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 8 / 26

  10. The Data WSJ Standard Corpus: WSJ k The Wall Street Journal section of the Penn Treebank Project (Marcus et al., 1993) ◮ ... stripped of punctuation, etc. ◮ ... rid of sentences left with more than k POS tags; ◮ ... and converted to reference dependencies — { t ∗ } , using “head percolation rules” (Collins, 1999) . Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 8 / 26

  11. The Data WSJ Standard Corpus: WSJ k The Wall Street Journal section of the Penn Treebank Project (Marcus et al., 1993) ◮ ... stripped of punctuation, etc. ◮ ... rid of sentences left with more than k POS tags; ◮ ... and converted to reference dependencies — { t ∗ } , using “head percolation rules” (Collins, 1999) . Training: traditionally, WSJ10 (Klein, 2005) ; Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 8 / 26

  12. The Data WSJ Standard Corpus: WSJ k The Wall Street Journal section of the Penn Treebank Project (Marcus et al., 1993) ◮ ... stripped of punctuation, etc. ◮ ... rid of sentences left with more than k POS tags; ◮ ... and converted to reference dependencies — { t ∗ } , using “head percolation rules” (Collins, 1999) . Training: traditionally, WSJ10 (Klein, 2005) ; Evaluation: Section 23 of WSJ ∞ (all sentences). Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 8 / 26

  13. The Data WSJ Standard Corpus: WSJ k 45 900 Sentences (1,000s) 40 800 35 700 30 600 25 500 20 400 15 300 10 200 Tokens (1,000s) 5 100 5 10 15 20 25 30 35 40 45 WSJ k Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 9 / 26

  14. The Data WSJ Standard Corpus: WSJ k 45 900 Sentences (1,000s) 40 800 35 700 30 600 25 500 20 400 15 300 10 200 Tokens (1,000s) 5 100 5 10 15 20 25 30 35 40 45 WSJ k Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 9 / 26

  15. Results Classic EM Classic EM: The Lay of the Land Directed Dependency Accuracy (%) on WSJ40 70 60 50 40 30 20 10 5 10 15 20 25 30 35 40 WSJ k

  16. Results Classic EM Classic EM: The Lay of the Land Directed Dependency Accuracy (%) on WSJ40 70 60 50 40 30 20 Uninformed 10 5 10 15 20 25 30 35 40 WSJ k

  17. Results Classic EM Classic EM: The Lay of the Land Directed Dependency Accuracy (%) on WSJ40 70 60 50 40 30 20 Uninformed 10 5 10 15 20 25 30 35 40 WSJ k

  18. Results Classic EM Classic EM: The Lay of the Land Directed Dependency Accuracy (%) on WSJ40 70 Oracle 60 50 40 30 20 Uninformed 10 5 10 15 20 25 30 35 40 WSJ k

  19. Results Classic EM Classic EM: The Lay of the Land Directed Dependency Accuracy (%) on WSJ40 70 60 50 Oracle 40 30 20 Uninformed 10 5 10 15 20 25 30 35 40 WSJ k

  20. Results Classic EM Classic EM: The Lay of the Land Directed Dependency Accuracy (%) on WSJ40 70 60 50 Oracle 40 30 20 Uninformed K&M ∗ 10 5 10 15 20 25 30 35 40 WSJ k

  21. Results Classic EM Classic EM: The Lay of the Land Directed Dependency Accuracy (%) on WSJ40 70 60 50 Oracle 40 30 20 Uninformed K&M ∗ 10 5 10 15 20 25 30 35 40 WSJ k Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 10 / 26

  22. Results Viterbi EM Viterbi EM: Results! Directed Dependency Accuracy (%) on WSJ40 70 Oracle 60 50 40 30 20 10 5 10 15 20 25 30 35 40 WSJ k

  23. Results Viterbi EM Viterbi EM: Results! Directed Dependency Accuracy (%) on WSJ40 70 Oracle 60 50 40 30 20 Uninformed 10 5 10 15 20 25 30 35 40 WSJ k

  24. Results Viterbi EM Viterbi EM: Results! Directed Dependency Accuracy (%) on WSJ40 70 Oracle 60 50 40 30 20 Uninformed K&M ∗ 10 5 10 15 20 25 30 35 40 WSJ k Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 11 / 26

  25. Results Viterbi EM State-of-the-Art Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 12 / 26

  26. Results Viterbi EM State-of-the-Art Section 23 of WSJ ∞ Right-Branching Baseline 32% (Klein and Manning, 2004) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 12 / 26

  27. Results Viterbi EM State-of-the-Art Section 23 of WSJ ∞ Right-Branching Baseline 32% (Klein and Manning, 2004) DMV with Classic EM 34% (Klein and Manning, 2004) 45% (Spitkovsky et al., 2010) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 12 / 26

  28. Results Viterbi EM State-of-the-Art Section 23 of WSJ ∞ Right-Branching Baseline 32% (Klein and Manning, 2004) DMV with Classic EM 34% (Klein and Manning, 2004) 45% (Spitkovsky et al., 2010) DMV with Viterbi EM with Smoothing 45% Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 12 / 26

  29. Results Viterbi EM State-of-the-Art Section 23 of WSJ ∞ Right-Branching Baseline 32% (Klein and Manning, 2004) DMV with Classic EM 34% (Klein and Manning, 2004) 45% (Spitkovsky et al., 2010) DMV with Viterbi EM with Smoothing 45% + Clever Initialization 48% Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 12 / 26

  30. Results Viterbi EM State-of-the-Art Section 23 of WSJ ∞ Brown100 Right-Branching Baseline 32% (Klein and Manning, 2004) DMV with Classic EM 34% (Klein and Manning, 2004) 45% 43% (Spitkovsky et al., 2010) DMV with Viterbi EM with Smoothing 45% 48% + Clever Initialization 48% 51% Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 12 / 26

  31. Results Viterbi EM State-of-the-Art Section 23 of WSJ ∞ Brown100 Right-Branching Baseline 32% (Klein and Manning, 2004) DMV with Classic EM 34% (Klein and Manning, 2004) 45% 43% (Spitkovsky et al., 2010) DMV with Viterbi EM ( + 5%) with Smoothing 45% 48% + Clever Initialization 48% 51% Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 12 / 26

  32. Results Viterbi EM State-of-the-Art Section 23 of WSJ ∞ Brown100 Right-Branching Baseline 32% (Klein and Manning, 2004) DMV with Classic EM 34% (Klein and Manning, 2004) 45% 43% (Spitkovsky et al., 2010) DMV with Viterbi EM ( + 5%) with Smoothing 45% 48% ( + 3%) + Clever Initialization 48% 51% Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 12 / 26

  33. Interpretation Interpretation: Why Does Viterbi EM Work? Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 13 / 26

  34. Interpretation Interpretation: Why Does Viterbi EM Work? in theory, Viterbi is a quick-and-dirty approximation Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 13 / 26

  35. Interpretation Interpretation: Why Does Viterbi EM Work? in theory, Viterbi is a quick-and-dirty approximation in theory, Communism works... Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 13 / 26

  36. Interpretation Interpretation: Why Does Viterbi EM Work? in theory, Viterbi is a quick-and-dirty approximation in theory, Communism works... in practice, EM emulates supervised learning: Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 13 / 26

  37. Interpretation Interpretation: Why Does Viterbi EM Work? in theory, Viterbi is a quick-and-dirty approximation in theory, Communism works... in practice, EM emulates supervised learning: s → { t } = T ( s ) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 13 / 26

  38. Interpretation Interpretation: Why Does Viterbi EM Work? in theory, Viterbi is a quick-and-dirty approximation in theory, Communism works... in practice, EM emulates supervised learning: s → { t } = T ( s ) Classic EM: w t = P θ ( t | s ) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 13 / 26

  39. Interpretation Interpretation: Why Does Viterbi EM Work? in theory, Viterbi is a quick-and-dirty approximation in theory, Communism works... in practice, EM emulates supervised learning: s → { t } = T ( s ) Classic EM: w t = P θ ( t | s ) clearly, this is redistribution of wealth mass Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 13 / 26

  40. Interpretation Interpretation: Why Does Viterbi EM Work? in theory, Viterbi is a quick-and-dirty approximation in theory, Communism works... in practice, EM emulates supervised learning: s → { t } = T ( s ) Classic EM: w t = P θ ( t | s ) clearly, this is redistribution of wealth mass — also, resembles an omniscient central planner Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 13 / 26

  41. Interpretation Interpretation: Why Does Viterbi EM Work? in theory, Viterbi is a quick-and-dirty approximation in theory, Communism works... in practice, EM emulates supervised learning: s → { t } = T ( s ) Classic EM: w t = P θ ( t | s ) clearly, this is redistribution of wealth mass — also, resembles an omniscient central planner (knows the true value of everything at all times) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 13 / 26

  42. Interpretation Interpretation: Why Does Viterbi EM Work? in theory, Viterbi is a quick-and-dirty approximation in theory, Communism works... in practice, EM emulates supervised learning: s → { t } = T ( s ) Classic EM: w t = P θ ( t | s ) clearly, this is redistribution of wealth mass — also, resembles an omniscient central planner (knows the true value of everything at all times) — could work, given a very powerful model θ ... Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 13 / 26

  43. Interpretation Interpretation: How Does Classic EM Fail? Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 14 / 26

  44. Interpretation Interpretation: How Does Classic EM Fail? our model is quite weak (e.g., doesn’t handle agreement) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 14 / 26

  45. Interpretation Interpretation: How Does Classic EM Fail? our model is quite weak (e.g., doesn’t handle agreement) reserves a lot of mass for ludicrous parse trees... Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 14 / 26

  46. Interpretation Interpretation: How Does Classic EM Fail? our model is quite weak (e.g., doesn’t handle agreement) reserves a lot of mass for ludicrous parse trees... — each entitled to non-trivial support by the distribution Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 14 / 26

  47. Interpretation Interpretation: How Does Classic EM Fail? our model is quite weak (e.g., doesn’t handle agreement) reserves a lot of mass for ludicrous parse trees... — each entitled to non-trivial support by the distribution at small scales, this is not a problem (short sentences) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 14 / 26

  48. Interpretation Interpretation: How Does Classic EM Fail? our model is quite weak (e.g., doesn’t handle agreement) reserves a lot of mass for ludicrous parse trees... — each entitled to non-trivial support by the distribution at small scales, this is not a problem (short sentences) — only so many possible parses → few free-loaders Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 14 / 26

  49. Interpretation Interpretation: How Does Classic EM Fail? our model is quite weak (e.g., doesn’t handle agreement) reserves a lot of mass for ludicrous parse trees... — each entitled to non-trivial support by the distribution at small scales, this is not a problem (short sentences) — only so many possible parses → few free-loaders eventually, exponentially many trees (unwashed masses) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 14 / 26

  50. Interpretation Interpretation: How Does Classic EM Fail? our model is quite weak (e.g., doesn’t handle agreement) reserves a lot of mass for ludicrous parse trees... — each entitled to non-trivial support by the distribution at small scales, this is not a problem (short sentences) — only so many possible parses → few free-loaders eventually, exponentially many trees (unwashed masses) result: a dog of a probability distribution... Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 14 / 26

  51. Interpretation Interpretation: How Does Classic EM Fail? our model is quite weak (e.g., doesn’t handle agreement) reserves a lot of mass for ludicrous parse trees... — each entitled to non-trivial support by the distribution at small scales, this is not a problem (short sentences) — only so many possible parses → few free-loaders eventually, exponentially many trees (unwashed masses) result: a dog of a probability distribution... Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 14 / 26

  52. Interpretation Interpretation: How Does Classic EM Fail? our model is quite weak (e.g., doesn’t handle agreement) reserves a lot of mass for ludicrous parse trees... — each entitled to non-trivial support by the distribution at small scales, this is not a problem (short sentences) — only so many possible parses → few free-loaders eventually, exponentially many trees (unwashed masses) result: a dog of a probability distribution... ... wagged by its very long tail Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 14 / 26

  53. Interpretation Interpretation: Idealogical Difference! Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 15 / 26

  54. Interpretation Interpretation: Idealogical Difference! Viterbi EM is powered by greed (much like Capitalism) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 15 / 26

  55. Interpretation Interpretation: Idealogical Difference! Viterbi EM is powered by greed (much like Capitalism) does not require ability to properly value all parse trees Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 15 / 26

  56. Interpretation Interpretation: Idealogical Difference! Viterbi EM is powered by greed (much like Capitalism) does not require ability to properly value all parse trees so long as it can spot a decent one (winner-take-all) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 15 / 26

  57. Interpretation Interpretation: Idealogical Difference! Viterbi EM is powered by greed (much like Capitalism) does not require ability to properly value all parse trees so long as it can spot a decent one (winner-take-all) different (weaker?) requirement on models: (like IR) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 15 / 26

  58. Interpretation Interpretation: Idealogical Difference! Viterbi EM is powered by greed (much like Capitalism) does not require ability to properly value all parse trees so long as it can spot a decent one (winner-take-all) different (weaker?) requirement on models: (like IR) — θ needs to be just discriminative enough! (ranking) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 15 / 26

  59. Interpretation Interpretation: Idealogical Difference! Viterbi EM is powered by greed (much like Capitalism) does not require ability to properly value all parse trees so long as it can spot a decent one (winner-take-all) different (weaker?) requirement on models: (like IR) — θ needs to be just discriminative enough! (ranking) at small scales, data are too sparse (markets are illiquid) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 15 / 26

  60. Interpretation Interpretation: Idealogical Difference! Viterbi EM is powered by greed (much like Capitalism) does not require ability to properly value all parse trees so long as it can spot a decent one (winner-take-all) different (weaker?) requirement on models: (like IR) — θ needs to be just discriminative enough! (ranking) at small scales, data are too sparse (markets are illiquid) improves with more data (statistics become efficient) Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 15 / 26

  61. Interpretation Interpretation: Idealogical Difference! Viterbi EM is powered by greed (much like Capitalism) does not require ability to properly value all parse trees so long as it can spot a decent one (winner-take-all) different (weaker?) requirement on models: (like IR) — θ needs to be just discriminative enough! (ranking) at small scales, data are too sparse (markets are illiquid) improves with more data (statistics become efficient) — really, what we want from unsupervised learners! Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 15 / 26

  62. Interpretation Interpretation: Summary Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 16 / 26

  63. Interpretation Interpretation: Summary Viterbi EM: focus on the individual best parse trees Spitkovsky et al. (Stanford & Google) Viterbi EM CoNLL (2010-07-15) 16 / 26

Recommend

Viterbi Training for PCFGs: Hardness Results and Competitiveness of Uniform Initialization Shay

Viterbi Training for PCFGs: Hardness Results and Competitiveness of Uniform Initialization Shay

Viterbi Training for PCFGs: Hardness Results and Competitiveness of Uniform Initialization Shay Cohen Noah Smith Carnegie Mellon University July 14, 2010 Outline Hardness results for unsupervised learning of PCFGs Background and problem

602 views • 46 slides
Viterbi decoder on STI CELL processor Michal Blaek (blazem2@fel.cvut.cz) Viterbi algorithm

Viterbi decoder on STI CELL processor Michal Blaek (blazem2@fel.cvut.cz) Viterbi algorithm

IBM - CVUT Student Research Projects Viterbi decoder on STI CELL processor Michal Blaek (blazem2@fel.cvut.cz) Viterbi algorithm introduction Finds the most likely state trajectory given the Works in several steps HMM model and

274 views • 5 slides
Search and Decoding Lecture 16 CS 753 Instructor: Preethi Jyothi Recall Viterbi search Viterbi

Search and Decoding Lecture 16 CS 753 Instructor: Preethi Jyothi Recall Viterbi search Viterbi

Search and Decoding Lecture 16 CS 753 Instructor: Preethi Jyothi Recall Viterbi search Viterbi search finds the most probable path through a trellis of time on the X-axis and states on the Y-axis q F end end end end v 1 (2) =.32 v 2

934 views • 23 slides
Pre-training on high-resource speech recognition improves low-resource speech-to-text translation

Pre-training on high-resource speech recognition improves low-resource speech-to-text translation

Pre-training on high-resource speech recognition improves low-resource speech-to-text translation Sameer Bansal Herman Kamper Karen Livescu Adam Lopez Sharon Goldwater Current systems Spanish Audio: ? English text: 2 Current systems

967 views • 71 slides
Coding and decoding with convolutional codes. The Viterbi Algorithm. J.-M. Brossier 2008 J.-M.

Coding and decoding with convolutional codes. The Viterbi Algorithm. J.-M. Brossier 2008 J.-M.

Convolutional encoding Finite State Machine Channel models The Viterbi algorithm Coding and decoding with convolutional codes. The Viterbi Algorithm. J.-M. Brossier 2008 J.-M. Brossier Coding and decoding with convolutional codes. The

788 views • 61 slides
Coding and decoding with convolutional codes. The Viterbi Algorithm. J.-M. Brossier 2008 J.-M.

Coding and decoding with convolutional codes. The Viterbi Algorithm. J.-M. Brossier 2008 J.-M.

Convolutional encoding Finite State Machine Channel models The Viterbi algorithm Coding and decoding with convolutional codes. The Viterbi Algorithm. J.-M. Brossier 2008 J.-M. Brossier Coding and decoding with convolutional codes. The

1.37k views • 135 slides
SOBER LIVING ENVIRONMENT Presentation WELCOME! TREATMENT Longer Treatment = Improves Outcome

SOBER LIVING ENVIRONMENT Presentation WELCOME! TREATMENT Longer Treatment = Improves Outcome

SOBER LIVING ENVIRONMENT Presentation WELCOME! TREATMENT Longer Treatment = Improves Outcome WORKS. More Support = Improves Outcome RECOVERY IS SOBER LIVING = IMPROVES OUTCOME! POSSIBLE! FO RM S FORMS PRECOVERY RESIDENCE

640 views • 26 slides
Algorithms for NLP CS 11-711 Fall 2020 Lecture 8: Viterbi, discriminative sequence labeling,

Algorithms for NLP CS 11-711 Fall 2020 Lecture 8: Viterbi, discriminative sequence labeling,

Algorithms for NLP CS 11-711 Fall 2020 Lecture 8: Viterbi, discriminative sequence labeling, NER Emma Strubell Announcements Project 1 is due tomorrow! You may submit up to 3 days late (out of a budget of 5 total for the semester).

890 views • 88 slides
Viterbi Algorithm Saravanan Vijayakumaran sarva@ee.iitb.ac.in Department of Electrical

Viterbi Algorithm Saravanan Vijayakumaran sarva@ee.iitb.ac.in Department of Electrical

Viterbi Algorithm Saravanan Vijayakumaran sarva@ee.iitb.ac.in Department of Electrical Engineering Indian Institute of Technology Bombay October 30, 2015 1 / 33 Encoder State Diagram 1 + D 2 1 + D + D 2 G ( D ) = 1 + D 0/000 v ( 0 )

1.26k views • 33 slides
Viterbi Algorithm Saravanan Vijayakumaran sarva@ee.iitb.ac.in Department of Electrical

Viterbi Algorithm Saravanan Vijayakumaran sarva@ee.iitb.ac.in Department of Electrical

Viterbi Algorithm Saravanan Vijayakumaran sarva@ee.iitb.ac.in Department of Electrical Engineering Indian Institute of Technology Bombay October 30, 2014 1 / 9 Encoder State Diagram 1 + D 2 1 + D + D 2 G ( D ) = 1 + D 0/000 v ( 0 )

402 views • 9 slides
Reflect, Learn, Act Exploration of reflection and how it improves interactions with patients ,

Reflect, Learn, Act Exploration of reflection and how it improves interactions with patients ,

Reflect, Learn, Act Exploration of reflection and how it improves interactions with patients , peers and professionals on placement Learning Outcomes To understand the importance of reflection and how it improves interactions between

333 views • 16 slides
Hidden Markov Models: Decoding & Training Natural Language Processing CMSC 35100 April 24,

Hidden Markov Models: Decoding & Training Natural Language Processing CMSC 35100 April 24,

Hidden Markov Models: Decoding & Training Natural Language Processing CMSC 35100 April 24, 2003 Agenda Speech Recognition Hidden Markov Models Uncertain observations Recognition: Viterbi, Stack/A* Training the

592 views • 21 slides
DRONE-SUPPORTED SURVEYING Increases social inclusion Improves prospects of peace Increases

DRONE-SUPPORTED SURVEYING Increases social inclusion Improves prospects of peace Increases

DRONE-SUPPORTED SURVEYING Increases social inclusion Improves prospects of peace Increases economic opportunities Improves governance Promote environmental protection and resilience Trimble UX5HP DJI Phantom 4 Pro Parrot ANAFI Work UAV

782 views • 41 slides
5/30/2014 Yielding Positions Prone positioning improves VQ To Prone or Not to mismatch

5/30/2014 Yielding Positions Prone positioning improves VQ To Prone or Not to mismatch

5/30/2014 Yielding Positions Prone positioning improves VQ To Prone or Not to mismatch Prone? Prone positioning improves arterial oxygenation John H. Turnbull, MD Assistant Professor of Anesthesia and Perioperative Medicine

89 views • 6 slides
Miler s Club s Club Miler Promoting Fitness Why get involved? Why get involved?

Miler s Club s Club Miler Promoting Fitness Why get involved? Why get involved?

Miler s Club s Club Miler Promoting Fitness Why get involved? Why get involved? Great way to get physically fit! Improves focus & concentration Decreases risk of MANY health problems Improves mood Improves

215 views • 10 slides
Illegitimi non carborundum Ronald L. Rivest Viterbi Professor of EECS MIT, Cambridge, MA CRYPTO

Illegitimi non carborundum Ronald L. Rivest Viterbi Professor of EECS MIT, Cambridge, MA CRYPTO

Illegitimi non carborundum Ronald L. Rivest Viterbi Professor of EECS MIT, Cambridge, MA CRYPTO 2011 2011-08-15 1 Illegitimi non carborundum (Dont let the bastards grind you down!) Ronald L. Rivest Viterbi Professor of EECS MIT,

1.45k views • 121 slides
Lecture 11: Viterbi and Forward Algorithms Kai-Wei Chang CS @ University of Virginia

Lecture 11: Viterbi and Forward Algorithms Kai-Wei Chang CS @ University of Virginia

Lecture 11: Viterbi and Forward Algorithms Kai-Wei Chang CS @ University of Virginia kw@kwchang.net Couse webpage: http://kwchang.net/teaching/NLP16 CS6501 Natural Language Processing 1 Quiz 1 Quiz 1 30 25 20 15 10 5 0 [0-5]

1k views • 42 slides
Thoughts On Appropriate Technologies for Voting Ronald L. Rivest Viterbi Professor of EECS MIT,

Thoughts On Appropriate Technologies for Voting Ronald L. Rivest Viterbi Professor of EECS MIT,

Thoughts On Appropriate Technologies for Voting Ronald L. Rivest Viterbi Professor of EECS MIT, Cambridge, MA Princeton CITP E-voting Workshop 2012-11-01 1 Is Voting Keeping Up with Technology? We live in an age of marvelous

1.35k views • 99 slides
9: Viterbi Algorithm for HMM Decoding Machine Learning and Real-world Data Helen Yannakoudakis 1

9: Viterbi Algorithm for HMM Decoding Machine Learning and Real-world Data Helen Yannakoudakis 1

9: Viterbi Algorithm for HMM Decoding Machine Learning and Real-world Data Helen Yannakoudakis 1 Computer Laboratory University of Cambridge Lent 2018 1 Based on slides created by Simone Teufel Last session: estimating parameters of an HMM The

1.17k views • 41 slides
9: Viterbi Algorithm for HMM Decoding Machine Learning and Real-world Data Simone Teufel and Ann

9: Viterbi Algorithm for HMM Decoding Machine Learning and Real-world Data Simone Teufel and Ann

9: Viterbi Algorithm for HMM Decoding Machine Learning and Real-world Data Simone Teufel and Ann Copestake Computer Laboratory University of Cambridge Lent 2017 Last session: estimating parameters of an HMM The dishonest casino, dice edition

837 views • 36 slides
Fast Polarization for Processes with Memory Boaz Shuval and Ido Tal Andrew and Erna Viterbi

Fast Polarization for Processes with Memory Boaz Shuval and Ido Tal Andrew and Erna Viterbi

Fast Polarization for Processes with Memory Boaz Shuval and Ido Tal Andrew and Erna Viterbi Department of Electrical Engineering Technion Israel Institute of Technology Haifa, 32000, Israel June 2018 1/21 In this talk Setting :

548 views • 27 slides
Universal Polarization for Processes with Memory Boaz Shuval and Ido Tal Andrew and Erna Viterbi

Universal Polarization for Processes with Memory Boaz Shuval and Ido Tal Andrew and Erna Viterbi

Universal Polarization for Processes with Memory Boaz Shuval and Ido Tal Andrew and Erna Viterbi Department of Electrical Engineering Technion Israel Institute of Technology Haifa, 32000, Israel July 2019 1/16 Setting Communication

487 views • 31 slides
A Tile-based Parallel Viterbi Algorithm for Biological Sequence Alignment on GPU with CUDA Zhihui

A Tile-based Parallel Viterbi Algorithm for Biological Sequence Alignment on GPU with CUDA Zhihui

A Tile-based Parallel Viterbi Algorithm for Biological Sequence Alignment on GPU with CUDA Zhihui Du 1 , Zhaoming Yin 2 , David A. Bader 3 1, Department of Computer Science, Tsinghua University. 2, School of Software and Microelectronics, Peking

536 views • 22 slides
Increased Apolipoprotein A-I produc2on improves MS-like

Increased Apolipoprotein A-I produc2on improves MS-like

Increased Apolipoprotein A-I produc2on improves MS-like symptoms in mice Lidia Gardner, Ph.D., Associate Professor Department of Neurology University

672 views • 24 slides

More recommend