Patternsofrealuserqueries10 Compared with the simulated caption data, the real user queries we collected show distinct patterns, e.g. Potential applications include22 retrieving very specific images of complex scenes the users encountered before, or exploring inspiring images for23 creativecontentgeneration(e.g. Weconjecture that unidirectional and bidirectional encoders provide comparable contextual signals when28 encoding theper-turn query asasingle feature vector fordownstream modules.

"B-Blue D-Black" stands for blue background and black digit).

Modularity is a compelling solution to continual learning (CL), the problem of modeling sequences of related tasks.

Cooperative multi-agent reinforcement learning (MARL) has been proposed for multi-agent collaborations toaccomplish manychallenging tasks[1,2,3,4].

Cooperative multi-agent reinforcement learning (MARL) has been proposed for multi-agent collaborations toaccomplish manychallenging tasks[1,2,3,4].

Statistical learning theory is the foundation of machine learning, providing theoretical bounds for the risk of models learned from a (single) training set, assumed to issue from an unknown probability distribution.

Despite its popularity, self-training is well-believed to be unreliable and often leads to training instability.

JS is also a special case of a general familyofdivergences,thef-divergences[20].

Based on this observation, we adopt ageneralized version ofthe JensenShannon divergence for multiple distributions to encourage consistency around data points. Using this loss function, we show state-of-the-art results on both synthetic(CIFAR),andreal-world(e.g.WebVision)noisewithvaryingnoiserates.