Research map / Architectures and efficiency / Attention and transformers
Context-Driven Incremental Compression for Multi-Turn Dialogue Generation
Yeongseo Jung, Jaehyeok Kim, Eunseo Jung, Jiachuan Wang, Yongqi Zhang, Ka Chun Cheung, Simon See, Lei Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add incremental compression (ablation) | C-DIC · REALTALK (all-sessions) | perplexity: 25.527 -> 9.356 | Improved |
| add incremental compression (ablation) | C-DIC · REALTALK (all-sessions) | ROUGE-2: 0.018 -> 0.056 | Improved |
| add retrieval-aware truncated BPTT (ablation) | C-DIC · REALTALK (all-sessions) | perplexity: 12.295 -> 9.356 | Improved |
| add memory-based context threading (ablation) | C-DIC · REALTALK (all-sessions) | perplexity: 9.197 -> 9.356 | Hurt |
| add memory-based context threading (ablation) | C-DIC · REALTALK (all-sessions) | BLEU: 0.046 -> 0.069 | Improved |
| replace replacement write-back with EMA (beta 0.7) (ablation) | C-DIC · MSC session 5 | perplexity: 8.427 -> 9.929 | Hurt |
| replace replacement write-back with EMA (beta 0.3) (ablation) | C-DIC · MSC session 5 | perplexity: 8.427 -> 8.442 | No clear effect |
| replace replacement write-back with learned gating (ablation) | C-DIC · MSC session 5 | ROUGE-L: 0.160 -> 0.162 | No clear effect |
| cap retrieved slots at 6 (bounded retrieval) (ablation) | C-DIC · REALTALK | perplexity: 9.579 -> 9.534 | No clear effect |
| use closed-loop instead of teacher-forcing history | C-DIC · REALTALK (all-sessions) | perplexity: 9.556 -> 9.576 | No clear effect |
| apply ICAE incrementally to its own compressed outputs | ICAE · MSC | perplexity: 27.656 -> 513.774 | Hurt |
| replace full prompting with C-DIC retrieved latent memory | Llama-2-Chat-7B · Latency at 30 turns | total time: 7.66 s -> 3.26 s | Improved |
Closest papers by meaning
- StreamingDialogue: Prolonged Dialogue Learning via Long Context Compression with Minimal Losses (NeurIPS 2024 · similarity 0.90)
- Compressed Context Memory for Online Language Model Interaction (ICLR 2024 · similarity 0.88)
- Dialogue Without Limits: Constant-Sized KV Caches for Extended Response in LLMs (ICML 2025 · similarity 0.88)
- SeCom: On Memory Construction and Retrieval for Personalized Conversational Agents (ICLR 2025 · similarity 0.87)
- TA&AT: Enhancing Task-Oriented Dialog with Turn-Level Auxiliary Tasks and Action-Tree Based Scheduled Sampling (AAAI 2024 · similarity 0.86)
- ToolACE-MT: Non-Autoregressive Generation for Agentic Multi-Turn Interaction (ICLR 2026 · similarity 0.86)
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.
