Paper Tracking

An automated weekly digest. Every Monday an agent scans arxiv and HuggingFace Daily Papers, downloads the PDFs, and summarizes the top 3 picks per topic.

Week of May 18, 2026 · View archive →

Graph × LLM

Graph-structured reasoning, graph foundation models, and LLM-augmented GNNs.

Decoupled and Divergence-Conditioned Prompt for Multi-domain Dynamic Graph Foundation Models

↑ 0 📚 136813 ★ 0 May 13

Problem: Multi-domain dynamic graphs have incompatible temporal semantics and divergent patterns, causing negative transfer in unified pre-training and adaptation.
Model: DyGFM: Dynamic Graph Foundation Model with semantic-temporal decoupled pre-training, divergence-aware expert routing, and divergence-conditioned prompt generation.
Code: RingBDStack/DyGFM

GFMate: Empowering Graph Foundation Models with Test-time Prompt Tuning

↑ 0 📚 13731 ★ 0 May 14

Problem: Existing graph foundation model prompts entangle source domain information and pre-training strategies, limiting generalization to unseen target domains and neglecting unlabelled test data.
Model: GFMate: pre-training-agnostic test-time graph prompt tuning using centroid and layer prompts with complementary learning objective
Code: YanJiangJerry/GFMate

A Unified Graph Language Model for Multi-Domain Multi-Task Graph Alignment Instruction Tuning

↑ 0 📚 23386 ★ 0 May 12

Problem: Existing graph language models lack unified alignment of multi-domain, multi-task GNN representations with LLM token spaces.
Model: "UniGraphLM: Unified Graph Language Model" with graph-text pair pretraining and curriculum alignment tuning for multi-domain, multi-task graph representation alignment with LLMs.
Code: not released

Vision-Language Models

Multimodal VLMs, vision encoders, and image-text foundation models.

MMSkills: Towards Multimodal Skills for General Visual Agents

↑ 99 📚 30599 ★ 0 May 13

Problem: Visual agents need reusable multimodal procedural knowledge that binds actions to visual state recognition and decision-making, beyond text-only skills.
Model: "MMSkills: framework for representing, generating, and utilizing reusable multimodal procedures for visual agents. Each skill couples textual procedures with runtime state cards and multi-view keyframes, generated via trajectory-to-skill generator and consulted via branch loading.
Code: DeepExperience/MMSkills

MemLens: Benchmarking Multimodal Long-Term Memory in Large Vision-Language Models

↑ 71 📚 3089 ★ 0 May 13

Problem: No benchmark systematically compares long-context LVLMs and memory-augmented agents on multimodal multi-session conversations requiring visual evidence.
Model: "MemLens": benchmark with 789 questions across five memory abilities (information extraction, multi-session reasoning, temporal reasoning, knowledge update, answer refusal) at four context lengths (32K-256K tokens)
Code: xrenaf/MEMLENS

MemEye: A Visual-Centric Evaluation Framework for Multimodal Agent Memory

↑ 58 📚 1421 ★ 0 May 13

Problem: Existing multimodal agent memory evaluations fail to assess whether agents preserve fine-grained visual evidence needed for reasoning over time.
Model: "MemEye: a visual-centric evaluation framework measuring visual evidence granularity and reasoning complexity in multimodal agent memory"
Code: not released
⚠ Interested, but agent could not fetch the PDF — summary based on abstract only.

World Models

Predictive world models, JEPA-style learning, video generation, and embodied simulators.

Causal Forcing++: Scalable Few-Step Autoregressive Diffusion Distillation for Real-Time Interactive Video Generation

↑ 87 📚 36802 ★ 0 May 14

Problem: Existing AR diffusion distillation methods require 4+ sampling steps; frame-wise 1–2 step generation needs efficient, scalable student initialization.
Model: "Causal Forcing++": AR diffusion distillation pipeline using causal consistency distillation for few-step student initialization, avoiding expensive full PF-ODE trajectory precomputation.
Code: thu-ml/Causal-Forcing

Warp-as-History: Generalizable Camera-Controlled Video Generation from One Training Video

↑ 38 📚 18942 ★ 0 May 13

Problem: Existing camera-controlled video generation methods require large-scale camera-annotated data or expensive test-time optimization; no simple way to leverage pretrained video models' latent camera-control capability.
Model: "Warp-as-History": converts camera-induced geometric warps into camera-warped pseudo-history fed through pretrained video models' native history pathway, with target-frame positional alignment and visible-token selection.
Code: yyfz/Warp-as-History

PhyMotion: Structured 3D Motion Reward for Physics-Grounded Human Video Generation

↑ 7 📚 19915 ★ 0 May 13

Problem: Existing video rewards fail to reliably score human motion realism because they operate in 2D pixel space without explicitly modeling 3D body physics and constraints.
Model: "PhyMotion": physics-grounded motion reward that recovers SMPL meshes, retargets to MuJoCo simulator, and evaluates motion via three axes (kinematic plausibility, contact/balance, dynamic feasibility)
Code: not released

Spatial Single-Cell Study

Single-cell + spatial omics foundation models, virtual cells, biological world models, and biomedical digital twins.

DUET: Dual-Paradigm Adaptive Expert Triage with Single-cell Inductive Prior for Spatial Transcriptomics Prediction

↑ 0 📚 5699 ★ 0 May 13

Problem: Existing methods for inferring spatial gene expression from histology images oversimplify morphology-to-expression mapping and underutilize large-scale single-cell data as biological constraints.
Model: DUET: dual-paradigm framework synergizing parametric regression and memory-based retrieval with cellular inductive priors and adaptive expert triage for spatial transcriptomics prediction.
Code: Junchao-Zhu/DUET

AssayBench: An Assay-Level Virtual Cell Benchmark for LLMs and Agents

↑ 0 📚 2120 ★ 0 May 11

Problem: No standard benchmark exists for predicting phenotypic outcomes of cellular perturbations across diverse assays in drug discovery.
Model: approach: LLM and agentic systems evaluated on gene ranking prediction from free-text screen descriptions using adjusted nDCG metric
Code: Genentech/AssayBench

StateXDiff: Cell State-Contextualized Multimodal Diffusion for Single-Cell Perturbation Prediction

↑ 0 ★ 0 May 15

Problem: Predicting drug-induced cellular state changes at single-cell resolution under out-of-distribution conditions with limited multimodal information.
Model: StateXDiff: cell State-contextualized multimodal Diffusion framework integrating transcriptomic and pseudo-protein representations with mechanism-aware drug templates via latent conditional diffusion.
Code: not released