Research map / Architectures and efficiency / Attention and transformers
Deconvolutional Paragraph Representation Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace recurrent decoding with deconvolutional decoding | CNN-DCNN autoencoder · Hotel Reviews | BLEU-4: 18.3 -> 94.2 | Helped |
| replace recurrent decoding with deconvolutional decoding | CNN-DCNN denoising autoencoder · Yahoo! Answers | CER: 0.2035 -> 0.1323 | Helped |
| replace recurrent decoding with deconvolutional decoding | CNN-DCNN denoising autoencoder · ArXiv | WER: 0.3819 -> 0.3067 | Helped |
| add joint reconstruction training to the classifier (ablation) | CNN text classifier · DBpedia | test error rate (%): 1.76 -> 1.17 | Helped |
| add joint reconstruction training to the classifier (ablation) | CNN text classifier · Yelp Review Polarity | test error rate (%): 4.62 -> 3.96 | Helped |
| add joint reconstruction training to the classifier (ablation) | CNN text classifier · Yahoo! Answers | test error rate (%): 27.42 -> 25.82 | Helped |
| use a deconvolutional rather than LSTM decoder during joint training (ablation) | Semi-supervised CNN classifier · DBpedia | test error rate (%): 1.36 -> 1.17 | Helped |
| use a deconvolutional rather than LSTM decoder during joint training (ablation) | Semi-supervised CNN classifier · Yelp Review Polarity | test error rate (%): 4.21 -> 3.96 | Helped |
| use a deconvolutional rather than LSTM decoder during joint training (ablation) | Semi-supervised CNN classifier · Yahoo! Answers | test error rate (%): 26.32 -> 25.82 | Helped |
| add unlabeled abstracts through joint training (ablation) | LSTM summarizer · ArXiv summarization, 100% observed | ROUGE-L: 16.37 -> 18.14 | Helped |
| add unlabeled abstracts through joint training (ablation) | LSTM summarizer · ArXiv summarization, 5% observed | ROUGE-L: 12.40 -> 16.04 | Helped |
| replace LSTM title decoder with deconvolutional decoder (ablation) | Semi-supervised summarizer · ArXiv summarization, 100% observed | ROUGE-L: 18.14 -> 16.83 | Hurt |
Closest papers by meaning
- Deconvolutional Latent-Variable Model for Text Sequence Matching (AAAI 2018 · similarity 0.89)
- Educating Text Autoencoders: Latent Representation Guidance via Denoising (ICML 2020 · similarity 0.87)
- Reading Scene Text in Deep Convolutional Sequences (AAAI 2016 · similarity 0.87)
- Recurrent Hierarchical Topic-Guided RNN for Language Generation (ICML 2020 · similarity 0.86)
- Fast Decoding in Sequence Models Using Discrete Latent Variables (ICML 2018 · similarity 0.85)
- Fast and Fluent Diffusion Language Models via Convolutional Decoding and Rejective Fine-tuning (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.