Feng Gao

Xiangming Xi, Feng Gao, Jun Xu et al.

Multi-task learning (MTL) is a paradigm that simultaneously learns multiple tasks by sharing information at different levels, enhancing the performance of each individual task. While previous research has primarily focused on feature-level or parameter-level task relatedness, and proposed various model architectures and learning algorithms to improve learning performance, we aim to explore output-level task relatedness. This approach introduces a posteriori information into the model, considering that different tasks may produce correlated outputs with mutual influences. We achieve this by incorporating a feedback mechanism into MTL models, where the output of one task serves as a hidden feature for another task, thereby transforming a static MTL model into a dynamic one. To ensure the training process converges, we introduce a convergence loss that measures the trend of a task's outputs during each iteration. Additionally, we propose a Gumbel gating mechanism to determine the optimal projection of feedback signals. We validate the effectiveness of our method and evaluate its performance through experiments conducted on several baseline models in spoken language understanding.

Sirui Xie, Feng Gao, Song-Chun Zhu

In searching for a generalizable representation of temporally extended tasks, we spot two necessary constituents: the utility needs to be non-Markovian to transfer temporal relations invariant to a probability shift, the utility also needs to be lifted to abstract out specific grounding objects. In this work, we study learning such utility from human demonstrations. While inverse reinforcement learning (IRL) has been accepted as a general framework of utility learning, its fundamental formulation is one concrete Markov Decision Process. Thus the learned reward function does not specify the task independently of the environment. Going beyond that, we define a domain of generalization that spans a set of planning problems following a schema. We hence propose a new quest, Generalized Inverse Planning, for utility learning in this domain. We further outline a computational framework, Maximum Entropy Inverse Planning (MEIP), that learns non-Markovian utility and associated concepts in a generative manner. The learned utility and concepts form a task representation that generalizes regardless of probability shift or structural change. Seeing that the proposed generalization problem has not been widely studied yet, we carefully define an evaluation protocol, with which we illustrate the effectiveness of MEIP on two proof-of-concept domains and one challenging task: learning to fold from demonstrations.