Open main menu

Wikipedia β

Multi-task learning (MTL) is a subfield of machine learning in which multiple learning tasks are solved at the same time, while exploiting commonalities and differences across tasks. This can result in improved learning efficiency and prediction accuracy for the task-specific models, when compared to training the models separately.[1][2][3] Early versions of MTL were called "hints"[4][5]

In a widely cited 1997 paper, Rich Caruana gave the following characterization:

Multitask Learning is an approach to inductive transfer that improves generalization by using the domain information contained in the training signals of related tasks as an inductive bias. It does this by learning tasks in parallel while using a shared representation; what is learned for each task can help other tasks be learned better.[3]

In the classification context, MTL aims to improve the performance of multiple classification tasks by learning them jointly. One example is a spam-filter, which can be treated as distinct but related classification tasks across different users. To make this more concrete, consider that different people have different distributions of features which distinguish spam emails from legitimate ones, for example an English speaker may find that all emails in Russian are spam, not so for Russian speakers. Yet there is a definite commonality in this classification task across users, for example one common feature might be text related to money transfer. Solving each user's spam classification problem jointly via MTL can let the solutions inform each other and improve performance.[6] Further examples of settings for MTL include multiclass classification and multi-label classification.[7]

Multi-task learning works because regularization induced by requiring an algorithm to perform well on a related task can be superior to regularization that prevents overfitting by penalizing all complexity uniformly. One situation where MTL may be particularly helpful is if the tasks share significant commonalities and are generally slightly under sampled.[6] However, as discussed below, MTL has also been shown to be beneficial for learning unrelated tasks.[8]

Contents

MethodsEdit

Task grouping and overlapEdit

Within the MTL paradigm, information can be shared across some or all of the tasks. Depending on the structure of task relatedness, one may want to share information selectively across the tasks. For example, tasks may be grouped or exist in a hierarchy, or be related according to some general metric. Suppose, as developed more formally below, that the parameter vector modeling each task is a linear combination of some underlying basis. Similarity in terms of this basis can indicate the relatedness of the tasks. For example with sparsity, overlap of nonzero coefficients across tasks indicates commonality. A task grouping then corresponds to those tasks lying in a subspace generated by some subset of basis elements, where tasks in different groups may be disjoint or overlap arbitrarily in terms of their bases.[9] Task relatedness can be imposed a priori or learned from the data.[7][10]. Hierarchical task relatedness can also be exploited implicitly without assuming a priori knowledge or learning relations explicitly.[11]

Exploiting unrelated tasksEdit

One can attempt learning a group of principal tasks using a group of auxiliary tasks, unrelated to the principal ones. In many applications, joint learning of unrelated tasks which use the same input data can be beneficial. The reason is that prior knowledge about task relatedness can lead to sparser and more informative representations for each task grouping, essentially by screening out idiosyncrasies of the data distribution. Novel methods which builds on a prior multitask methodology by favoring a shared low-dimensional representation within each task grouping have been proposed. The programmer can impose a penalty on tasks from different groups which encourages the two representations to be orthogonal. Experiments on synthetic and real data have indicated that incorporating unrelated tasks can result in significant improvements over standard multi-task learning methods.[8]

Transfer of knowledgeEdit

Related to multi-task learning is the concept of knowledge transfer. Whereas traditional multi-task learning implies that a shared representation is developed concurrently across tasks, transfer of knowledge implies a sequentially shared representation. Large scale machine learning projects such as the deep convolutional neural network GoogLeNet,[12] an image-based object classifier, can develop robust representations which may be useful to further algorithms learning related tasks. For example, the pre-trained model can be used as a feature extractor to perform pre-processing for another learning algorithm. Or the pre-trained model can be used to initialize a model with similar architecture which is then fine-tuned to learn a different classification task.[13]

Group online adaptive learningEdit

Traditionally Multi-task learning and transfer of knowledge are applied to stationary learning settings. Their extension to non-stationary environments is termed Group online adaptive learning (GOAL).[14] Sharing information could be particularly useful if learners operate in continuously changing environments, because a learner could benefit from previous experience of another learner to quickly adapt to their new environment. Such group-adaptive learning has numerous applications, from predicting financial time-series, through content recommendation systems, to visual understanding for adaptive autonomous agents.

MathematicsEdit

Reproducing Hilbert space of vector valued functions (RKHSvv)Edit

The MTL problem can be cast within the context of RKHSvv (a complete inner product space of vector-valued functions equipped with a reproducing kernel). In particular, recent focus has been on cases where task structure can be identified via a separable kernel, described below. The presentation here derives from Ciliberto et al, 2015.[7]

RKHSvv conceptsEdit

Suppose the training data set is  , with  ,  , where   indexes task, and  . Let  . In this setting there is a consistent input and output space and the same loss function   for each task: . This results in the regularized machine learning problem:

 

 

 

 

 

(1)

where   is a vector valued reproducing kernel Hilbert space with functions   having components  .

The reproducing kernel for the space   of functions   is a symmetric matrix-valued function   , such that   and the following reproducing property holds:

 

 

 

 

 

(2)

The reproducing kernel gives rise to a representer theorem showing that any solution to equation 1 has the form:

 

 

 

 

 

(3)

Separable kernelsEdit

The form of the kernel   induces both the representation of the feature space and structures the output across tasks. A natural simplification is to choose a separable kernel, which factors into separate kernels on the input space   and on the tasks  . In this case the kernel relating scalar components   and   is given by  . For vector valued functions   we can write  , where   is a scalar reproducing kernel, and   is a symmetric positive semi-definite   matrix. Henceforth denote   .

This factorization property, separability, implies the input feature space representation does not vary by task. That is, there is no interaction between the input kernel and the task kernel. The structure on tasks is represented solely by  . Methods for non-separable kernels   is an current field of research.

For the separable case, the representation theorem is reduced to  . The model output on the training data is then   , where   is the   empirical kernel matrix with entries  , and   is the   matrix of rows  .

With the separable kernel, equation 1 can be rewritten as

 

 

 

 

 

(P)

where   is a (weighted) average of   applied entry-wise to Y and KCA. (The weight is zero if   is a missing observation).

Note the second term in P can be derived as follows:

 

  (bilinearity)

  (reproducing property)

 

Known task structureEdit

Task structure representationsEdit

There are three largely equivalent ways to represent task structure: through a regularizer; through an output metric, and through an output mapping.

Regularizer - With the separable kernel, it can be shown (below) that  , where   is the   element of the pseudoinverse of  , and   is the RKHS based on the scalar kernel  , and  . This formulation shows that   controls the weight of the penalty associated with  . (Note that   arises from  .)

Proof:

 

 

 

 

 

 

 

 

 

Output metric - an alternative output metric on   can be induced by the inner product  . With the squared loss there is an equivalence between the separable kernels   under the alternative metric, and  , under the canonical metric.

Output mapping - Outputs can be mapped as   to a higher dimensional space to encode complex structures such as trees, graphs and strings. For linear maps  , with appropriate choice of separable kernel, it can be shown that  .

Task structure examplesEdit

Via the regularizer formulation, one can represent a variety of task structures easily.

  • Letting   (where   is the TxT identity matrix, and   is the TxT matrix of ones) is equivalent to letting   control the variance   of tasks from their mean  . For example, blood levels of some biomarker may be taken on   patients at   time points during the course of a day and interest may lie in regularizing the variance of the predictions across patients.
  • Letting   , where   is equivalent to letting   control the variance measured with respect to a group mean:  . (Here   the cardinality of group r, and   is the indicator function). For example people in different political parties (groups) might be regularized together with respect to predicting the favorability rating of a politician. Note that this penalty reduces to the first when all tasks are in the same group.
  • Letting  , where   is the Laplacian for the graph with adjacency matrix M giving pairwise similarities of tasks. This is equivalent to giving a larger penalty to the distance separating tasks t and s when they are more similar (according to the weight  ,) i.e.   regularizes  .
  • All of the above choices of A also induce the additional regularization term   which penalizes complexity in f more broadly.

Learning tasks together with their structureEdit

Learning problem P can be generalized to admit learning task matrix A as follows:

 

 

 

 

 

(Q)

Choice of   must be designed to learn matrices A of a given type. See "Special cases" below.

Optimization of QEdit

Restricting to the case of convex losses and coercive penalties Ciliberto et al have shown that although Q is not convex jointly in C and A, a related problem is jointly convex.

Specifically on the convex set  , the equivalent problem

 

 

 

 

 

(R)

is convex with the same minimum value. And if   is a minimizer for R then   is a minimizer for Q.

R may be solved by a barrier method on a closed set by introducing the following perturbation:

 

 

 

 

 

(S)

The perturbation via the barrier   forces the objective functions to be equal to   on the boundary of   .

S can be solved with a block coordinate descent method, alternating in C and A. This results in a sequence of minimizers   in S that converges to the solution in R as  , and hence gives the solution to Q.

Special casesEdit

Spectral penalties - Dinnuzo et al[15] suggested setting F as the Frobenius norm  . They optimized Q directly using block coordinate descent, not accounting for difficulties at the boundary of  .

Clustered tasks learning - Jacob et al[16] suggested to learn A in the setting where T tasks are organized in R disjoint clusters. In this case let   be the matrix with  . Setting  , and  , the task matrix   can be parameterized as a function of  :   , with terms that penalize the average, between clusters variance and within clusters variance respectively of the task predictions. M is not convex, but there is a convex relaxation  . In this formulation,  .

GeneralizationsEdit

Non-convex penalties - Penalties can be constructed such that A is constrained to be a graph Laplacian, or that A has low rank factorization. However these penalties are not convex, and the analysis of the barrier method proposed by Ciliberto et al does not go through in these cases.

Non-separable kernels - Separable kernels are limited, in particular they do not account for structures in the interaction space between the input and output domains jointly. Future work is needed to develop models for these kernels.

ApplicationsEdit

Spam filteringEdit

Using the principles of MTL, techniques for collaborative spam filtering that facilitates personalization have been proposed. In large scale open membership email systems, most users do not label enough messages for an individual local classifier to be effective, while the data is too noisy to be used for a global filter across all users. A hybrid global/individual classifier can be effective at absorbing the influence of users who label emails very diligently from the general public. This can be accomplished while still providing sufficient quality to users with few labeled instances.[17]

Web searchEdit

Using boosted decision trees, one can enable implicit data sharing and regularization. This learning method can be used on web-search ranking data sets. One example is to use ranking data sets from several countries. Here, multitask learning is particularly helpful as data sets from different countries vary largely in size because of the cost of editorial judgments. It has been demonstrated that learning various tasks jointly can lead to significant improvements in performance with surprising reliability.[18]

RoboEarthEdit

In order to facilitate transfer of knowledge, IT infrastructure is being developed. One such project, RoboEarth, aims to set up an open source internet database that can be accessed and continually updated from around the world. The goal is to facilitate a cloud-based interactive knowledge base, accessible to technology companies and academic institutions, which can enhance the sensing, acting and learning capabilities of robots and other artificial intelligence agents.[19]

Software packageEdit

The Multi-Task Learning via StructurAl Regularization (MALSAR) Matlab package [20] implements the following multi-task learning algorithms:

  • Mean-Regularized Multi-Task Learning[21][22]
  • Multi-Task Learning with Joint Feature Selection[23]
  • Robust Multi-Task Feature Learning[24]
  • Trace-Norm Regularized Multi-Task Learning[25]
  • Alternating Structural Optimization[26][27]
  • Incoherent Low-Rank and Sparse Learning[28]
  • Robust Low-Rank Multi-Task Learning
  • Clustered Multi-Task Learning[29][30]
  • Multi-Task Learning with Graph Structures

See alsoEdit

ReferencesEdit

  1. ^ Baxter, J. (2000). A model of inductive bias learning" Journal of Artificial Intelligence Research 12:149--198, On-line paper
  2. ^ Thrun, S. (1996). Is learning the n-th thing any easier than learning the first?. In Advances in Neural Information Processing Systems 8, pp. 640--646. MIT Press. Paper at Citeseer
  3. ^ a b Caruana, R. (1997). "Multi-task learning" (PDF). Machine Learning. 28: 41–75. doi:10.1023/A:1007379606734. 
  4. ^ Suddarth, S., Kergosien, Y. (1990). Rule-injection hints as a means of improving network performance and learning time. EURASIP Workshop. Neural Networks pp. 120-129. Lecture Notes in Computer Science. Springer.
  5. ^ Abu-Mostafa, Y. S. (1990). "Learning from hints in neural networks". Journal of Complexity. 6: 192–198. doi:10.1016/0885-064x(90)90006-y. 
  6. ^ a b Weinberger, Kilian. "Multi-task Learning". 
  7. ^ a b c Ciliberto, C. (2015). "Convex Learning of Multiple Tasks and their Structure". arXiv:1504.03101 . 
  8. ^ a b Romera-Paredes, B., Argyriou, A., Bianchi-Berthouze, N., & Pontil, M., (2012) Exploiting Unrelated Tasks in Multi-Task Learning. http://jmlr.csail.mit.edu/proceedings/papers/v22/romera12/romera12.pdf
  9. ^ Kumar, A., & Daume III, H., (2012) Learning Task Grouping and Overlap in Multi-Task Learning. http://icml.cc/2012/papers/690.pdf
  10. ^ Jawanpuria, P., & Saketha Nath, J., (2012) A Convex Feature Learning Formulation for Latent Task Structure Discovery. http://icml.cc/2012/papers/90.pdf
  11. ^ Zweig, A. & Weinshall, D. Hierarchical Regularization Cascade for Joint Learning. Proceedings: of 30th International Conference on Machine Learning (ICML), Atlanta GA, June 2013. http://www.cs.huji.ac.il/~daphna/papers/Zweig_ICML2013.pdf
  12. ^ Szegedy, C. (2014). "Going Deeper with Convolutions". Computer Vision and Pattern Recognition (CVPR), 2015 IEEE Conference on. arXiv:1409.4842 . doi:10.1109/CVPR.2015.7298594. 
  13. ^ Roig, Gemma. "Deep Learning Overview" (PDF). 
  14. ^ Zweig, A. & Chechik, G. Group online adaptive learning. Machine Learning, DOI 10.1007/s10994-017- 5661-5, August 2017. http://rdcu.be/uFSv
  15. ^ Dinuzzo, Francesco (2011). "Learning output kernels with block coordinate descent.". Proceedings of the 28th International Conference on Machine Learning (ICML-11). 
  16. ^ Jacob, Laurent (2009). "Clustered multi-task learning: A convex formulation". Advances in neural information processing systems. 
  17. ^ Attenberg, J., Weinberger, K., & Dasgupta, A. Collaborative Email-Spam Filtering with the Hashing-Trick. http://www.cse.wustl.edu/~kilian/papers/ceas2009-paper-11.pdf
  18. ^ Chappelle, O., Shivaswamy, P., & Vadrevu, S. Multi-Task Learning for Boosting with Application to Web Search Ranking. http://www.cse.wustl.edu/~kilian/papers/multiboost2010.pdf
  19. ^ Description of RoboEarth Project
  20. ^ Zhou, J., Chen, J. and Ye, J. MALSAR: Multi-tAsk Learning via StructurAl Regularization. Arizona State University, 2012. http://www.public.asu.edu/~jye02/Software/MALSAR. On-line manual
  21. ^ Evgeniou, T., & Pontil, M. (2004). Regularized multi–task learning. Proceedings of the tenth ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 109–117).
  22. ^ Evgeniou, T.; Micchelli, C.; Pontil, M. (2005). "Learning multiple tasks with kernel methods" (PDF). Journal of Machine Learning Research. 6: 615. 
  23. ^ Argyriou, A.; Evgeniou, T.; Pontil, M. (2008a). "Convex multi-task feature learning". Machine Learning. 73: 243–272. doi:10.1007/s10994-007-5040-8. 
  24. ^ Chen, J., Zhou, J., & Ye, J. (2011). Integrating low-rank and group-sparse structures for robust multi-task learning. Proceedings of the tenth ACM SIGKDD international conference on Knowledge discovery and data mining.
  25. ^ Ji, S., & Ye, J. (2009). An accelerated gradient method for trace norm minimization. Proceedings of the 26th Annual International Conference on Machine Learning (pp. 457–464).
  26. ^ Ando, R.; Zhang, T. (2005). "A framework for learning predictive structures from multiple tasks and unlabeled data". The Journal of Machine Learning Research. 6: 1817–1853. 
  27. ^ Chen, J., Tang, L., Liu, J., & Ye, J. (2009). A convex formulation for learning shared structures from multiple tasks. Proceedings of the 26th Annual International Conference on Machine Learning (pp. 137–144).
  28. ^ Chen, J., Liu, J., & Ye, J. (2010). Learning incoherent sparse and low-rank patterns from multiple tasks. Proceedings of the 16th ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 1179–1188).
  29. ^ Jacob, L., Bach, F., & Vert, J. (2008). Clustered multi-task learning: A convex formulation. Advances in Neural Information Processing Systems, 2008
  30. ^ Zhou, J., Chen, J., & Ye, J. (2011). Clustered multi-task learning via alternating structure optimization. Advances in Neural Information Processing Systems.

External linksEdit