An EM Approach to Learning Sequential

We consider problems of sequence processing and we propose a solution based on a discrete state model. We introduce a recurrent architecture having a modular structure that allocates subnetworks to discrete states. Di erent subnetworks are model the dynamics (state transition) and the output of the model, conditional on the previous state and an external input. The model has a statistical interpretation and can be trained by the EM or GEM algorithms, considering state trajectories as missing data. This allows to decouple temporal credit assignment and actual parameters estimation. The model presents similarities to hidden Markov models, but allows to map input sequences to output sequences, using the same processing style of recurrent networks. For this reason we call it Input/Output HMM (IOHMM). Another remarkable di erence is that IOHMMs are trained using a supervised learning paradigm (while potentially taking advantage of the EM algorithm), whereas standard HMMs are trained by an unsupervised EM algorithm (or a supervised criterion with gradient ascent). We also study the problem of learning long-term dependencies with Markovian systems, making comparisons to recurrent networks trained by gradient descent. The analysis reported in this paper shows that Markovian models generally su er from a problem of di usion of temporal credit for long-term dependencies and fully connected transition graphs. However, while recurrent networks exhibit a con ict between long-term information storing and trainability, these two requirements are either both satis ed or both not satis ed in Markovian models. Finally, we demonstrate that EM supervised learning is well suited for solving grammatical inference problems. Experimental results are presented for the seven Tomita grammars, showing that these adaptive models can attain excellent generalization. 1