6arXiv cs.LG (Machine Learning)·
Seriality gap identified as structural limitation of video diffusion models on causal chain tasks
A new arXiv paper identifies and formalizes the 'seriality gap' in video diffusion models: a structural mismatch where tasks requiring growing serial computation (e.g., multi-ball physics chains) degrade as causal chain length increases, even with more denoising steps. Controlled experiments on hard-sphere dynamics isolate dependent-event structure—not video length—as the cause. The authors prove that for deterministic video prediction, denoising steps add no serial computation beyond the backbone, and show that autoregressive/blockwise generation and architectural depth help mitigate the gap. This has direct implications for using video diffusion as a world model or physics simulator.
Related guides (2)
Related events (8)
VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion
VideoMLA applies Multi-Head Latent Attention (MLA) to causal video diffusion, replacing per-head keys and values with a shared low-rank content latent and decoupled 3D-RoPE positional key, achieving 92.7% reduction in per-token KV memory. The paper investigates why MLA works despite pretrained video attention not being low-rank (unlike the spectral assumption motivating MLA in LLMs), finding that the MLA bottleneck itself determines effective rank rather than the pretrained spectrum. On VBench, VideoMLA matches short-horizon baselines, achieves best overall score at long horizons, and delivers 1.23x throughput improvement on a single NVIDIA B200 GPU.
Video generation models as world simulators
OpenAI introduces Sora, a large-scale text-conditional video diffusion model built on a transformer architecture that operates on spacetime patches of video and image latent codes. The model is trained jointly on videos and images of variable durations, resolutions, and aspect ratios. Sora can generate up to one minute of high-fidelity video and OpenAI frames scaling video generation as a path toward general-purpose physical world simulators.
Kolmogorov Regression lifts diffusion policies to Cameron-Martin space for robust long-horizon control
Researchers introduce a backward Kolmogorov equation framework that reformulates diffusion policy training as a deterministic boundary-value PDE problem in Cameron-Martin space, replacing stochastic score matching. The approach uses a precision-weighted Cameron-Martin loss and a Kolmogorov residual as an inference-time failure detector, yielding convergence guarantees tied to kernel effective rank rather than action dimension. Validation on the PushT manipulation benchmark shows 17% improvement in episode reward and 67.6% reduction in inter-step drift; a 6-station manufacturing scheduling task shows 28.4% lower RMSE than LSTM baselines and 96% reduction in deadlock events via Hamilton-Jacobi reachability certification.
Triplet-Block Diffusion RWKV: Unifying Linear-Time Causal Models with Bidirectional Discrete Diffusion
The paper introduces B³D-RWKV, a 7.2B-parameter language model that combines RWKV's O(L) linear-time inference with parallel bidirectional discrete diffusion via a triplet-block layout. This architecture resolves the fundamental tension between causal (unidirectional) and diffusion (bidirectional) attention requirements. On an 8-task evaluation suite, B³D-RWKV-7.2B achieves comparable accuracy to existing models while delivering an average 1.6× decoding throughput speedup over baselines.
TunerDiT: Training-free Progressive Steering of Diffusion Transformers for Multi-Event Video Generation
TunerDiT is a training-free method for steering video diffusion transformers (DiTs) to generate long-horizon videos containing multiple sequential events. The approach identifies intrinsic turning points in the DiT denoising trajectory where text conditioning shifts from global layout to fine-grained detail, then applies two steering mechanisms: Event-Partitioned Masking and Cross-Event Prompt Fusion. The authors also introduce Meve, a benchmark prompt suite for multi-event video generation, and report state-of-the-art results across 8 metrics with improved text alignment scaling with event count.
The Abstraction Gap in Vision-Language Causal Reasoning
Researchers introduce a dual-probe methodology and the CAGE benchmark (49,500 questions across 5,500 images) to distinguish linguistic plausibility from faithful causal reasoning in vision-language models. An Abstraction Gap (AG) metric quantifies the normalized performance difference between text-only and chain-of-reasoning probes. Evaluating eight VLMs, seven exhibit AG exceeding 0.50—generating fluent causal text but failing structured causal chain tasks—while one model achieves near-zero AG, suggesting architectural and pretraining choices are decisive. Fine-tuning on 45,000 chain-annotated examples fails to close the gap, pointing to a fundamental capability distinction.
Finite-Sample Lens for Understanding Diffusion Posterior Sampler Failures
This paper introduces a finite-sample theoretical framework for analyzing diffusion model posterior samplers used in imaging inverse problems. The authors show that popular likelihood approximations at intermediate timesteps systematically under- or over-estimate posterior spread, leading to failure modes including sensitivity to early stopping, incorrect weighting of posterior modes, and hallucination of prior or likelihood modes. Crucially, they demonstrate these failures can arise from a multimodal prior alone, without requiring nonlinear measurement models or multimodal posteriors. The framework is model-agnostic and can serve as a diagnostic tool for evaluating existing and future posterior samplers.
Phantom specialization in circuit discovery: structural differences don't imply distinct mechanisms
A new arXiv preprint challenges a core assumption in mechanistic interpretability: that structurally different circuits discovered for the same task imply distinct computational mechanisms. Using Literal Sequence Copying across token-frequency bands in five Pythia models (70M–1.4B), the authors extract 75 circuits and show that structurally distinct circuits implement the same computation, with band-specific edges transferring broadly and a shared core recovering ≥99% of circuit performance. The paper introduces the term 'phantom specialization' for this pattern and argues that standard source-level evaluation inflates apparent faithfulness, while edge-level evaluation and cross-condition transfer tests are needed to detect the many-to-one mapping from structure to function.