Research map / Probabilistic and scientific modeling / Variational and generative sampling
A Discrete Variational Recurrent Topic Model without the Reparametrization Trick
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 |
|---|---|---|---|
| Use VRTM-LSTM with explicit word-level topics instead of TGVAE | VRTM-LSTM versus TGVAE · APNEWS test | perplexity: $\leq$48.73 -> $\leq\mathbf{47.78}$ | Helped |
| Use VRTM-LSTM with explicit word-level topics instead of TGVAE | VRTM-LSTM versus TGVAE · IMDB test | perplexity: $\leq$57.11 -> $\leq\mathbf{51.08}$ | Helped |
| Use VRTM-LSTM with explicit word-level topics instead of TGVAE | VRTM-LSTM versus TGVAE · BNC test | perplexity: $\leq$87.86 -> $\leq\mathbf{86.33}$ | Helped |
| Increase the number of topics from 10 to 50 (ablation) | VRTM-LSTM · APNEWS, IMDB, and BNC test | perplexity: APNEWS: $\leq$54.31 -> $\leq\mathbf{47.78}$; IMDB: $\leq$59.82 -> $\leq\mathbf{51.08}$; BNC: $\leq$92.89 -> $\leq\mathbf | Helped |
| Use pretrained word2vec embeddings instead of embeddings learned from scratch (ablation) | VRTM-LSTM · APNEWS, T=50 | perplexity: 51.35 -> 52.31 | Hurt |
| Use VRTM instead of LDA VB | VRTM versus LDA VB · APNEWS test, 5 topics | SwitchP: 0.26 -> 0.59 | Helped |
| Use VRTM instead of LDA VB | VRTM versus LDA VB · APNEWS test, 50 topics | SwitchP: 0.08 -> 0.20 | Helped |
| Tune the Dirichlet prior parameter alpha (ablation) | VRTM variational topic distributions · 10-topic model | topic assignment entropy | Helped |
Closest papers by meaning
- Discovering Discrete Latent Topics with Neural Variational Inference (ICML 2017 · similarity 0.94)
- Neural Dynamic Focused Topic Model (AAAI 2023 · similarity 0.91)
- Neural Variational Inference for Text Processing (ICML 2016 · similarity 0.90)
- An Instability in Variational Inference for Topic Models (ICML 2019 · similarity 0.90)
- Autoencoding Variational Inference For Topic Models (ICLR 2017 · similarity 0.88)
- textTOvec: DEEP CONTEXTUALIZED NEURAL AUTOREGRESSIVE TOPIC MODELS OF LANGUAGE WITH DISTRIBUTED COMPOSITIONAL PRIOR (ICLR 2019 · similarity 0.88)
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.