Add SparsePR to modality-temporal methods

consisency-paper/modality+temporal-consistency/README.md

@@ -1193,4 +1193,35 @@
 </details>
 </details>
 
----
+---
+
+<details>
+<summary><b>Partition the Support, Reconstruct the Residual: Training-Free Sparse Attention for Video Generation and World Models</b></summary>
+
+* **Authors:** Pardis Taghavi, Reza Langari, Gaurav Pandey  
+* **arXiv ID:** 2608.18484  
+* **One-liner:** A training-free block-sparse attention method that couples shared support routing with probe-fitted residual reconstruction for video generators and world models.  
+* **Published in:** arXiv 2026
+* **Links:** [[Paper]](https://arxiv.org/abs/2608.18484) | [[PDF]](https://arxiv.org/pdf/2608.18484.pdf) | [[Code]](https://github.com/PardisTaghavi/SparsePR) | [[Project Page]](https://pardistaghavi.github.io/SparsePR-website/)
+
+> **Core Innovation**  
+> Combines Response-Coupled Partitioning, which groups sampled-query key responses to produce shared block routes, with Probe-Fitted Residual Reconstruction, which uses a small set of exact query rows to calibrate a call-specific affine correction for skipped interactions.
+
+<details>
+    <summary>Abstract</summary>
+    Training-free block-sparse attention can accelerate video transformers, but row-wise attention concentration does not by itself specify an executable sparse operator. Queries sharing a block route may have poorly overlapping supports, while retained attention mass alone does not determine the post-softmax error from skipped interactions. We show that partition geometry affects both pooled support and the predictability of the remaining residual from the sparse output.
+
+    We introduce SparsePR, which combines Response-Coupled Partitioning with Probe-Fitted Residual Reconstruction. Sampled-query key responses form paired K/V groups, whose centroids induce query-response coordinates for shared routing. A small set of exact query rows then calibrates a call-specific affine correction from the sparse output within the output subspace observed in the probe residuals.
+
+    Across four heterogeneous video generation and world models, SparsePR consistently reduces attention-reconstruction error. Ablations show that probe fitting accounts for most of this reduction, while response-coupled partitioning lowers hard-drop error and improves reconstruction under a finite probe budget. SparsePR preserves generation quality at 22.0-26.0% realized executed-pair density while achieving 1.48x-2.61x end-to-end speedups.
+</details>
+
+<details>
+    <summary>Key points</summary>
+    * **Response-Coupled Partitioning:** Forms paired K/V response groups whose centroids define shared sparse routing coordinates.  
+    * **Probe-Fitted Reconstruction:** Uses a small number of exact query rows to estimate a call-specific correction for skipped attention interactions.  
+    * **Practical Efficiency:** Preserves generation quality at 22.0-26.0% executed-pair density and reports 1.48x-2.61x end-to-end speedups across four video generation and world models.  
+</details>
+</details>
+
+---