Benchmark

Weekly scoreboard of top papers by Signal Fusion. Data is updated weekly by our pipeline.

Week of 2026-03-30 · v2

Rank	Paper	Viability	Composite	Rank Δ
1	Adaptive Block-Scaled Data Types	9.0	122.9	—
2	Gen-Searcher: Reinforcing Agentic Search for Image Generation	9.0	122.6	—
3	CiQi-Agent: Aligning Vision, Tools and Aesthetics in Multimodal Agent for Cultural Reasoning on Chinese Porcelains	8.0	113.7	—
4	DreamLite: A Lightweight On-Device Unified Model for Image Generation and Editing	8.0	112.1	—
5	AdaptToken: Entropy-based Adaptive Token Selection for MLLM Long Video Understanding	8.0	111.0	—
6	MarkushGrapher-2: End-to-end Multimodal Recognition of Chemical Structures	8.0	110.7	—
7	Kernel-Smith: A Unified Recipe for Evolutionary Kernel Optimization	8.0	110.6	—
8	DRIVE-Nav: Directional Reasoning, Inspection, and Verification for Efficient Open-Vocabulary Navigation	7.0	103.1	—
9	See it to Place it: Evolving Macro Placements with Vision-Language Models	7.0	102.4	—
10	PoseDreamer: Scalable and Photorealistic Human Data Generation Pipeline with Diffusion Models	7.0	99.7	—
11	Ghost-FWL: A Large-Scale Full-Waveform LiDAR Dataset for Ghost Detection and Removal	8.0	80.0	—
12	ORACAL: A Robust and Explainable Multimodal Framework for Smart Contract Vulnerability Detection with Causal Graph Enrichment	8.0	80.0	—
13	SVGS: Single-View to 3D Object Editing via Gaussian Splatting	8.0	80.0	—
14	$AutoDrive\text{-}P^3$: Unified Chain of Perception-Prediction-Planning Thought via Reinforcement Fine-Tuning	8.0	80.0	—
15	Event6D: Event-based Novel Object 6D Pose Tracking	8.0	80.0	—
16	Scaling Atomistic Protein Binder Design with Generative Pretraining and Test-Time Compute	8.0	80.0	—
17	AMIGO: Agentic Multi-Image Grounding Oracle Benchmark	7.0	70.0	—
18	On-the-fly Repulsion in the Contextual Space for Rich Diversity in Diffusion Transformers	7.0	70.0	—
19	RehearsalNeRF: Decoupling Intrinsic Neural Fields of Dynamic Illuminations for Scene Editing	7.0	70.0	—
20	FlowIt: Global Matching for Optical Flow with Confidence-Guided Refinement	7.0	70.0	—
21	EpiScreen: Early Epilepsy Detection from Electronic Health Records with Large Language Models	7.0	70.0	—
22	Trust-Aware Routing for Distributed Generative AI Inference at the Edge	7.0	70.0	—
23	Navigating the Mirage: A Dual-Path Agentic Framework for Robust Misleading Chart Question Answering	7.0	70.0	—
24	Hydra: Unifying Document Retrieval and Generation in a Single Vision-Language Model	7.0	70.0	—
25	SonoWorld: From One Image to a 3D Audio-Visual Scene	7.0	70.0	—
26	Subspace Optimization for Backpropagation-Free Continual Test-Time Adaptation	7.0	70.0	—
27	Seeing with You: Perception-Reasoning Coevolution for Multimodal Reasoning	7.0	70.0	—
28	EBuddy: a workflow orchestrator for industrial human-machine collaboration	7.0	70.0	—
29	Rethinking Language Model Scaling under Transferable Hypersphere Optimization	7.0	70.0	—
30	FL-PBM: Pre-Training Backdoor Mitigation for Federated Learning	7.0	70.0	—
31	TGIF2: Extended Text-Guided Inpainting Forgery Dataset & Benchmark	7.0	70.0	—
32	ChemCLIP: Bridging Organic and Inorganic Anticancer Compounds Through Contrastive Learning	7.0	70.0	—
33	FocusVLA: Focused Visual Utilization for Vision-Language-Action Models	7.0	70.0	—
34	LACE: Loss-Adaptive Capacity Expansion for Continual Learning	7.0	70.0	—
35	Learning Partial Action Replacement in Offline MARL	7.0	70.0	—
36	Bit-Identical Medical Deep Learning via Structured Orthogonal Initialization	7.0	70.0	—
37	RAD-AI: Rethinking Architecture Documentation for AI-Augmented Ecosystems	7.0	70.0	—
38	Industrial3D: A Terrestrial LiDAR Point Cloud Dataset and CrossParadigm Benchmark for Industrial Infrastructure	7.0	70.0	—
39	ResAdapt: Adaptive Resolution for Efficient Multimodal Reasoning	7.0	70.0	—
40	CirrusBench: Evaluating LLM-based Agents Beyond Correctness in Real-World Cloud Service Environments	7.0	70.0	—
41	Pandora: Articulated 3D Scene Graphs from Egocentric Vision	7.0	70.0	—
42	Divide and Restore: A Modular Task-Decoupled Framework for Universal Image Restoration	7.0	70.0	—
43	Unsafe2Safe: Controllable Image Anonymization for Downstream Utility	7.0	70.0	—
44	XSPA: Crafting Imperceptible X-Shaped Sparse Adversarial Perturbations for Transferable Attacks on VLMs	7.0	70.0	—
45	SAGAI-MID: A Generative AI-Driven Middleware for Dynamic Runtime Interoperability	7.0	70.0	—
46	Mitigating Backdoor Attacks in Federated Learning Using PPA and MiniMax Game Theory	7.0	70.0	—
47	ELViS: Efficient Visual Similarity from Local Descriptors that Generalizes Across Domains	7.0	70.0	—
48	StreamingVLA: Streaming Vision-Language-Action Model with Action Flow Matching and Adaptive Early Observation	7.0	70.0	—
49	HandX: Scaling Bimanual Motion and Interaction Generation	7.0	70.0	—
50	SOLE-R1: Video-Language Reasoning as the Sole Reward for On-Robot Reinforcement Learning	7.0	70.0	—

Freshness + Provenance

Last updated: 2026-03-30
Source count: 1
Coverage window: Week of 2026-03-30
Method version: v2

Sources: benchmark_snapshots, papers, predictions, metric_snapshots

Methodology

Rankings use Signal Fusion: (viability_score × 10) + (unicorn_probability × 30) + (votes × 0.1) + (star_velocity × 5). See our FAQ for full scoring definition.