M.A. Marcinkiewicz’s scientific contributions

... In this section, we validate the performance and the generality of the proposed LSbased Hyperpruning (LSH) method across various RNN architectures and datasets. We evaluate stacked LSTM and RHN models on the Penn TreeBank Dataset [83], and AWD-LSTM-MoS on the WikiText-2 Dataset [84] for language modeling tasks. To ensure comparability, we adopt the same experimental setup as in prior work [53] for consistency. ...

Reference:

Hyperpruning: Efficient Search through Pruned Variants of Recurrent Neural Networks Leveraging Lyapunov Spectrum

... Our approach has been tested separately on three different datasers including ATIS (Hemphill et al., 1990), IBM Manual and Wall Street Journal (WSJ) (Marcus et al., 1993) corpora. Some parts of these datasets were randomly selected and divided into three parts: initialization dataset (IDS), training dataset (TRDS) and testing dataset (TSDS). ...

Reference:

Bridge the Gap between Statistical and Hand-crafted Grammars