S. W. Campbell

Niall O' Mahony, Sean Campbell, Lenka Krpalkova et al.

Fine-Grained Change Detection and Regression Analysis are essential in many applications of ArtificialIntelligence. In practice, this task is often challenging owing to the lack of reliable ground truth information andcomplexity arising from interactions between the many underlying factors affecting a system. Therefore,developing a framework which can represent the relatedness and reliability of multiple sources of informationbecomes critical. In this paper, we investigate how techniques in multi-task metric learning can be applied for theregression of fine-grained change in real data.The key idea is that if we incorporate the incremental change in a metric of interest between specific instancesof an individual object as one of the tasks in a multi-task metric learning framework, then interpreting thatdimension will allow the user to be alerted to fine-grained change invariant to what the overall metric isgeneralised to be. The techniques investigated are specifically tailored for handling heterogeneous data sources,i.e. the input data for each of the tasks might contain missing values, the scale and resolution of the values is notconsistent across tasks and the data contains non-independent and identically distributed (non-IID) instances. Wepresent the results of our initial experimental implementations of this idea and discuss related research in thisdomain which may offer direction for further research.

Rohit Prabhavalkar, Yanzhang He, David Rybach et al.

End-to-end models that condition the output label sequence on all previously predicted labels have emerged as popular alternatives to conventional systems for automatic speech recognition (ASR). Since unique label histories correspond to distinct models states, such models are decoded using an approximate beam-search process which produces a tree of hypotheses. In this work, we study the influence of the amount of label context on the model's accuracy, and its impact on the efficiency of the decoding process. We find that we can limit the context of the recurrent neural network transducer (RNN-T) during training to just four previous word-piece labels, without degrading word error rate (WER) relative to the full-context baseline. Limiting context also provides opportunities to improve the efficiency of the beam-search process during decoding by removing redundant paths from the active beam, and instead retaining them in the final lattice. This path-merging scheme can also be applied when decoding the baseline full-context model through an approximation. Overall, we find that the proposed path-merging scheme is extremely effective allowing us to improve oracle WERs by up to 36% over the baseline, while simultaneously reducing the number of model evaluations by up to 5.3% without any degradation in WER.