Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.06831 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913171848036352 |
|---|---|
| author | Ashiq, Muhammad H. Arora, Samanyu Harish, Abhinav N. Kharbanda, Ishaan Tseng, Hung Yun Chrysos, Grigorios G. |
| author_facet | Ashiq, Muhammad H. Arora, Samanyu Harish, Abhinav N. Kharbanda, Ishaan Tseng, Hung Yun Chrysos, Grigorios G. |
| contents | We theoretically study the hallucination phenomena in two canonical diffusion samplers: the stochastic Denoising Diffusion Probabilistic Model (DDPM) and the deterministic Denoising Diffusion Implicit Model (DDIM). We analyze the reverse ODE (DDIM) and SDE (DDPM) for a Gaussian mixture target, proving that after a critical time $τ$, (a) DDIM can become stuck on the segment connecting the two nearest modes and (b) DDPM *stochasticity* helps it become unstuck from this region, thus avoiding hallucination. Our empirical validation verifies that DDPM has a significantly lower hallucination rate than DDIM when this region is entered. Building on our observations, we exhibit how using additional stochastic steps can help DDIM avoid hallucinations and offer new insights on how to design improved samplers. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_06831 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Why DDIM Hallucinates More Than DDPM: A Theoretical Analysis of Reverse Dynamics Ashiq, Muhammad H. Arora, Samanyu Harish, Abhinav N. Kharbanda, Ishaan Tseng, Hung Yun Chrysos, Grigorios G. Machine Learning Artificial Intelligence We theoretically study the hallucination phenomena in two canonical diffusion samplers: the stochastic Denoising Diffusion Probabilistic Model (DDPM) and the deterministic Denoising Diffusion Implicit Model (DDIM). We analyze the reverse ODE (DDIM) and SDE (DDPM) for a Gaussian mixture target, proving that after a critical time $τ$, (a) DDIM can become stuck on the segment connecting the two nearest modes and (b) DDPM *stochasticity* helps it become unstuck from this region, thus avoiding hallucination. Our empirical validation verifies that DDPM has a significantly lower hallucination rate than DDIM when this region is entered. Building on our observations, we exhibit how using additional stochastic steps can help DDIM avoid hallucinations and offer new insights on how to design improved samplers. |
| title | Why DDIM Hallucinates More Than DDPM: A Theoretical Analysis of Reverse Dynamics |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2605.06831 |