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.
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.
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. Further examples of settings for MTL include multiclass classification and multi-label classification.
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. However, as discussed below, MTL has also been shown to be beneficial for learning unrelated tasks.
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. Task relatedness can be imposed a priori or learned from the data.
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.
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, 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.
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:
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:
The reproducing kernel gives rise to a representer theorem showing that any solution to equation 1 has the form:
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
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:
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 .)
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 .
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 structure
Learning problem P can be generalized to admit learning task matrix A as follows:
Choice of must be designed to learn matrices A of a given type. See "Special cases" below.
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
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:
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.
Spectral penalties - Dinnuzo et al 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 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, .
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.
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.
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.
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.
^Argyriou, A., Evgeniou, T., & Pontil, M. (2008a). Convex multi-task feature learning. Machine Learning, 73, 243–272.
^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.
^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).
^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.
^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).
^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).
^Jacob, L., Bach, F., & Vert, J. (2008). Clustered multi-task learning: A convex formulation. Advances in Neural Information Processing Systems， 2008
^Zhou, J., Chen, J., & Ye, J. (2011). Clustered multi-task learning via alternating structure optimization. Advances in Neural Information Processing Systems.