Download presentation
1
Learning from labelled and unlabeled data
Semi-Supervised Learning Machine Learning – PDEEC 2008/2009 Filipe Tiago Alves de Magalhães
2
Unsupervised Learning
Semi-Supervised Learning Supervised Learning Semi-Supervised Learning Unsupervised Learning Labbeled + unlabeled data discover patterns in the data that relate data attributes with a target (class) attribute. The data have no target attribute (unlabeled). Typically, plenty of unlabeled data available. We want to explore the data to find some intrinsic structures in them. These patterns are then utilized to predict the values of the target attribute in future data instances. Tries to improve the predictive power using both labelled and unlabeled data. (Expected to be better than using one alone)
3
Semi-Supervised Learning
Unlabeled data is easy to obtain Labelled data can be difficult to obtain - human annotation is boring - may require experts - may require special equipment - very time-consuming Examples: Web page classification (billions of pages) classification (SPAM or No-SPAM) Speech annotation (400h for each hour of conversation) …
4
Semi-Supervised Learning
Semi-Supervised learning can be seen as an excellent way to improve the results that we would get using exclusively supervised or non-supervised methods, for the same scenario. Although we (or specialists) do not need to spend such a big effort labelling data, a great concern must be faced for the design of good models, feature extraction, kernels definition.
5
Semi-Supervised Learning
Sometimes, it may not be so hard to label data… Tries to guess the user’s gender based on his/her choices. After that, we tell if it was right or wrong Takes advantage of player’s intervention in order to enrich the training of automatic learning algorithms
6
7th IEEE Workshops on Application of Computer Vision (WACV/MOTION'05)
Semi-Supervised Self-Training of Object Detection Models Chuck Rosenberg Google, Inc. Martial Hebert Carnegie Mellon University Henry Schneiderman Carnegie Mellon University 7th IEEE Workshops on Application of Computer Vision (WACV/MOTION'05) 2005
7
Semi-Supervised Learning Self-Training
L = (Xi , Yi ) Set of labelled data U = (Xi , ? ) Set of unlabeled data Algorithm Repeat Train a classifier C with training data L Classify data in U with C Find a subset U’ of U with the most confident scores L + U’ L U – U’ U
8
Semi-Supervised Self-Training of Object Detection Models
Object detection based on its shape - time-consuming - exhaustive labelling (background, foreground, object, non-object) Try to simplify the collection and preparation of training data - combining data labelled in different ways - labelling of each image region can take the form of a probability distribution over labels (“weakly” labelled) - e.g., is more likely that the object is present in the centre of the image - e.g., a certain image has a high likelihood of containing the object, but its position is unknown.
9
If Semi-Supervised Self-Training of Object Detection Models
Training Approaches Generic detection algorithm for classification of a subwindow in an image as being part of the “object” class or the “clutter/everything else” class If X – image feature vectors xi – data at a specific location in the image (i = {1, … ,n} indexes images locations) Y – class f – foreground b – background θf – parameters of the foreground model θb – parameters of the background model
10
Semi-Supervised Self-Training of Object Detection Models
Training Approaches EM approach
11
Semi-Supervised Self-Training of Object Detection Models
Training Approaches EM approach There are many reasons why EM may not perform well in a particular semi-supervised training context. EM solely finds a set of model parameters which maximize the likelihood of the data. - Fully labeled data may not sufficiently constrain the solution, which means that there may be solutions which maximize the data likelihood but do not optimize classification performance.
12
Semi-Supervised Self-Training of Object Detection Models
Training Approaches Alternative
13
Semi-Supervised Self-Training of Object Detection Models
Detector Overview (Experimental Setup) Subwindow is processed for lighting correction Two-level wavelet transform is applied Features are computed by vector quantizing groups of wavelet coefficients Subwindow is classified by thresholding a linear combination of the log-likelihood ratios of the features Cascade architecture → only image patches which are accepted by the first detector are passed on to the next
14
Semi-Supervised Self-Training of Object Detection Models
Data (Experimental Setup) Landmark used on a typical training image sample training images and the training examples associated with them Set with positive examples – 231 images 480 training examples Independent test set – 44 images 102 test examples 15000 negative examples Training examples – 24 x 16 pixels (rotated, scaled and cropped) pixels high and pixels wide
15
Semi-Supervised Self-Training of Object Detection Models
Training (Experimental Setup) Training the model with fully labeled data consists of the following steps: Given the training data landmark locations geometrically normalize the training example subimages; apply lighting normalization to the subimages; generate synthetic training examples (scaling, shifting and rotating) Compute the wavelet transform of the subimages Quantize each group of wavelet coefficients and build a naïve Bayes model with respect to each group to discriminate between positive and negative examples Adjust the naïve Bayes model using boosting, but maintaining a linear decision function, effectively performing gradient descent on the margin Compute a ROC curve for the detector using a cross validation set Choose a threshold for the linear function, based on the final performance desired
16
Semi-Supervised Self-Training of Object Detection Models
Selection Metrics (Experimental Setup) Selection metric is crucial to the performance of the training Confidence selection Computed at every iteration by applying the detector trained from the current set of labelled data to the weakly labelled data set. Detection with highest confidence is selected and added to the training set MSE selection Is calculated for each weakly labelled example by evaluating the distance between the corresponding image window and all of the other templates in the training data (including the original labelled examples and the weakly labelled examples added in prior iterations)
17
Semi-Supervised Self-Training of Object Detection Models
Selection Metrics (Experimental Setup) The candidate image and the labeled images are first normalized with a specific set of processing steps before the MSE based score metric is computed. The score is based on the Mahalanobis distance
18
Semi-Supervised Self-Training of Object Detection Models
Selection Metrics (Experimental Setup) position Detector MSE selection metric scale The detector must be accurate in localization but need not be accurate in detection since false detection will be discarded due to their large MSE distances to all of the training examples. This is crucial to ensure the performance of the training algorithm with small initial training sets. This is also part of the reason for the MSE to outperform the confidence metric, which requires the detector to be accurate in both localization and detection performance.
19
Semi-Supervised Self-Training of Object Detection Models
Experiment Scenarios (Experiments and Analysis) Each experiment was repeated using a different initial random subset, in order to avoid the variance that was being observed in the detector performance and in the behaviour of the semi-supervised training process. Experiment = specific set of experimental conditions Run = each repetition of that experiment Mostly, 5 runs were performed for each experiment Typically, 20 weakly labelled images were added to the training set at each iteration, because of the substantial training time of the detector. Ideally, only a single image would be added at each iteration.
20
Semi-Supervised Self-Training of Object Detection Models
Evaluation Metrics (Experiments and Analysis) Each run was evaluated by using the area under the ROC curve (AUC). Because different experimental conditions affect performance, the AUCs were normalized relatively to the full data performance of that run. if (performance level = = 1.0) { the model being evaluated has the same performance as it would if all of the labelled data was utilised } if (performance level < 1.0) the model has a lower performance than that achieved with the full data set To compute the full data performance, each specific run is trained with the full data set and its performance is recorded.
21
Semi-Supervised Self-Training of Object Detection Models
Baseline training configurations (Experiments and Analysis) Smooth regime was chosen in order to perform experiments under conditions where the addition of weakly labelled data would make a difference.
22
Semi-Supervised Self-Training of Object Detection Models
Selection Metrics (Experiments and Analysis) Does the choice of the selection metric make a substantial difference in the performance of the semi-supervised training? Confidence metric MSE metric
23
Semi-Supervised Self-Training of Object Detection Models
Selection Metrics (Experiments and Analysis) Does the choice of the selection metric make a substantial difference in the performance of the semi-supervised training? decreases increases
24
Semi-Supervised Self-Training of Object Detection Models
Relative size of fully Labelled Data(Experiments and Analysis) How many weakly labelled examples do we need to add to the training set in order to reach the best detector performance?
25
Semi-Supervised Self-Training of Object Detection Models
Conclusions/Discussion The results showed that it was possible to achieve detection performance that was close to the base performance obtained with the fully labelled data, even when a small fraction of the training data was used in the initial training set. The experiments showed that the self-training approach to semi-supervised training can be applied to an existing detector that was originally designed for supervised training. The MSE selection metric consistently outperformed the confidence metric. More generally, the self-training approach using an independently-defined selection metric outperforms both the confidence metrics and the batch EM approaches. During the training process, the distribution of the labeled data at any particular iteration may not match the actual underlying distribution of the data.
26
Semi-Supervised Self-Training of Object Detection Models
Conclusions/Discussion True labels for the unlabeled data Original unlabeled data and labelled data (c),(d) The points labelled by the incremental self-training algorithm after 5 iterations using the confidence metric and the Euclidean metric, respectively.
27
Semi-Supervised Self-Training of Object Detection Models
Future Work Study the relation between the semi-supervised training approach evaluated here with the co-training approaches. Develop more precise guidelines for selecting the initial training set. The approach could be extended to training examples that are labelled in different ways. For example, some images may be provided with scale information and nothing else. Additional information may be provided such as the rough shape of the object, or a prior distribution over its location in the image.
28
ZZZZZZZZZZZZZZ….. Still Awake???
29
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization Andrew B. Goldberg Computer Sciences Department University of Wisconsin-Madison Xiaojin Zhu Computer Sciences Department University of Wisconsin-Madison TextGraphs: HLT/NAACL Workshop on Graph-based Algorithms for Natural Language Processing 2006
30
? ? ? Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization Sentiment Categorization ? ? ?
31
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization Sentiment Categorization
32
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization What we saw is rating inference Bo Pang and Lillian Lee Seeing stars: Exploiting class relationships for sentiment categorization with respect to rating scales. In Proceedings of the ACL. In this work… Graph-based Semi-supervised Learning Main assumption encoded in graph: Similar documents should have similar ratings
33
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization
34
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization
35
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization
36
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization
37
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization 50% accuracy
38
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization
39
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization 100% accuracy
40
Goal Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization Goal
41
Approach Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization Approach
42
Measuring Loss over the Graph
Seeing stars when there aren’t many stars: Graph-based semi-supervised learning for sentiment categorization Measuring Loss over the Graph
43
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization
44
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization
45
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization
46
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization
47
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization
48
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization
49
Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization Minimization now is non- trivial
50
Finding a Closed-Form Solution
Seeing stars when there aren’t many stars: Graph-based semi-supervised learning for sentiment categorization Finding a Closed-Form Solution Fortunately, we
51
Finding a Closed-Form Solution
Seeing stars when there aren’t many stars: Graph-based semi-supervised learning for sentiment categorization Vector of f values for all reviews Vector of given labels yi for labelled reviews and predicted labels for unlabeled reviews Finding a Closed-Form Solution C = Labelled Unlabeled
52
Graph Laplacian matrix
Seeing stars when there aren’t many stars: Graph-based semi-supervised learning for sentiment categorization Graph Laplacian matrix Finding a Closed-Form Solution Constant parameter
53
Graph Laplacian Matrix
Seeing stars when there aren’t many stars: Graph-based semi-supervised learning for sentiment categorization Graph Laplacian Matrix Assume n labelled and unlabeled documents
54
Finding a Closed-Form Solution
Seeing stars when there aren’t many stars: Graph-based semi-supervised learning for sentiment categorization Finding a Closed-Form Solution
55
Experiments Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization Experiments Predict 1 to 4 stars ratings for reviews 4-author data (Pang and Lee, 2005) 1770, 902, 1307 and 1027 documents, respectively Each document represented as a {0,1} word-presence vector, normalized to sum 1 Positive-Sentence Percentage (PSP) similarity (Pang and Lee, 2005) Tuned parameters with cross-validation * * Joachims, T., Transductive Inference for Text Classification using Support Vector Machines, in Proceedings of the Sixteenth International Conference on Machine Learning. 1999, Morgan Kaufmann Publishers Inc.
56
Experiments Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization Experiments PSPi is defined as the percentage of positive sentences in review xi. The similarity between reviews xi, xj is the cosine angle between the vectors (PSPi,1-PSPi) and (PSPj, 1-PSPj) Positive sentences are identified using a binary classifier trained on a “snippet data set” (10662 documents)
57
Experiments Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization Experiments Low ratings tend to get low PSP scores High ratings tend to get high PSP scores The trend was qualitatively the same as in Pang and Lee (2005) (Naïve Bayes)
58
α = ak + bk’ c = k/L Experiments
Seeing stars when there aren’t many stars: Graph-based semi-supervised learning for sentiment categorization Experiments Number of unlabeled neighbours α = ak + bk’ c = k/L Size of labelled set Number of labelled neighbours Optimal Values (through cross-validation) c = 0.2 α = 1.5
59
outperforms other methods for small labelled set sizes
Seeing stars when there aren’t many stars: Graph-based semi-supervised learning for sentiment categorization Results Graph-based SSL outperforms other methods for small labelled set sizes
60
Alternative Similarity Measure
Seeing stars when there aren’t many stars: Graph-based semi-supervised learning for sentiment categorization Alternative Similarity Measure The cosine between word vectors containing all words, each weighted by its mutual information Scaling of mutual information values (maximum = 1) Previously found values → weights for corresponding words in the word vectors Words in the movie review data that did not appear in the “snippet data set” were excluded Optimal Values (through cross-validation) c = 0.1 α = 1.5
61
Results Seeing stars when there aren’t many stars:
Graph-based semi-supervised learning for sentiment categorization Results In each row, in green is the best result and any results that could not be distinguished from it with a paired t-test at the 0.05 level. 20 trial average unlabeled set accuracy for each author across different labelled set sizes and methods
62
Conclusions and Future Work
Seeing stars when there aren’t many stars: Graph-based semi-supervised learning for sentiment categorization Conclusions and Future Work Graph-based semi-supervised learning based on PSP similarity achieved better performance than all other methods in all four author corpora. However, for larger labelled sets its performance was not so good. a) Maybe, because SVM regressor trained on a large labelled set can achieve fairly high __accuracy without considering relationships between examples. b) PSP similarity is not accurate enough, thus biasing the overall performance when labelled __data is abundant. Investigate better document representations and similarity measures. Extend the method to inductive learning setting Experiment cross-reviewer and cross-domain analysis, such as using a model learned on movie reviews to help classify product reviews.
63
Human Semi-Supervised Learning
Q: Do humans also use semi-supervised learning? A: Apparently, yes!
64
Hint: Temporal association
Human Semi-Supervised Learning Some evidences… Face recognition is a very challenging computational task. However, it is an easy task for humans. Differences between two views of the same face are much larger than those between two different faces viewed at the same angle. + + Sinha, P., et al., Face recognition by humans: 20 results all computer vision researchers should know about. 2006, MIT. Hint: Temporal association
65
Human Semi-Supervised Learning
Some evidences… Observers were shown sequences of novel faces in which the identity of the face changed as the head rotated. image sequence Unlabeled data As a result, observers showed a tendency to treat the views as if they were of the same person. suggests We are continuously associating views of objects to support later recognition, and that we do so not only on the basis of the physical similarity, but also the correlated appearance in time of the objects. Wallis, G. and H. Bülthoff, Effects of temporal association on recognition memory, in National Academy of Sciences p
66
Human Semi-Supervised Learning
Some evidences… 17-month infants listen to a word, see an object They wanted to measure their ability to associate the word and the object If the word was heard many times before (without seeing the object; unlabeled data), association was stronger. If the word was not heard before, association was weaker. Graf, E., et al., Can Infants Map Meaning to Newly Segmented Words?: Statistical Segmentation and Word Learning. Psychological Science, (3): p Image taken from
67
Human Semi-Supervised Learning
Better understanding of the human cognitive model, can guide the development of better machine learning algorithms or make existent even better and robust…
68
References Rosenberg, C., M. Hebert, and H. Schneiderman, Semi-Supervised Self-Training of Object Detection Models, in Proceedings of the Seventh IEEE Workshops on Application of Computer Vision (WACV/MOTION'05) - Volume 1 - Volume , IEEE Computer Society. Goldberg, A.B. and X. Zhu. Seeing stars when there aren't many stars: Graph-based semi-supervised learning for sentiment categorization. in TextGraphs: HLT/NAACL Workshop on Graph-based Algorithms for Natural Language Processing Bo Pang and Lillian Lee Seeing stars: Exploiting class relationships for sentiment categorization with respect to rating scales. In Proceedings of the ACL. Joachims, T., Transductive Inference for Text Classification using Support Vector Machines, in Proceedings of the Sixteenth International Conference on Machine Learning. 1999, Morgan Kaufmann Publishers Inc. Sinha, P., et al., Face recognition by humans: 20 results all computer vision researchers should know about. 2006, MIT. Wallis, G. and H. Bülthoff, Effects of temporal association on recognition memory, in National Academy of Sciences p Graf, E., et al., Can Infants Map Meaning to Newly Segmented Words?: Statistical Segmentation and Word Learning. Psychological Science, (3): p
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.