ML Reads

Personal arXiv list

ML papers to read today.

Pick a topic and keep a small daily list of papers worth opening.

Refresh queueDaily mix

Today's queue

5 papers

#01Nov 19, 2026

cs.CL

MemFolder: Structure-Guided Memory Folding for Efficient Agent Memory Management

Xuyang Liu, Chenwen Lu, Junxi Wang and 6 more

Long-term agents accumulate large external memory stores as interactions unfold, increasing the costs of storage, retrieval, and response generation. Efficiency is constrained by two coupled factors: a growing record count increases retrieval overhead, while longer retrieved contexts increase prefill computation and decoding-time attention costs. Existing methods prune or consolidate records, but pruning can discard useful evidence, whereas consolidation may produce fewer yet longer retrieval units. We introduce MemFolder, a structure-guided memory folding framework that sequentially reduces record count and context length. The framework comprises two core components that share memory groups, using cross-record structure to guide both evidence consolidation and compact representation. RecordFolder grounds entity mentions, progressively selects representative yet complementary anchors, and combines semantic and entity affinity with temporal chaining to consolidate compatible records into fewer retrieval units. ContextFolder encodes the original records within these groups into compact text and uses verification feedback to revise encodings when key facts cannot be recovered. Experiments on the three text-based memory benchmarks LoCoMo, PersonaMem, and LongMemEval and the two multimodal benchmarks M3-Bench-web and M3-Bench-robot demonstrate the strong performance of our methods. Notably, on Mem0, RecordFolder retains 99.0% of the original average performance across the three text-based benchmarks with 50% fewer records, exceeding the strongest baseline by 1.9 percentage points in performance retention. Combined with ContextFolder, existing record compression methods further shorten their output records while retaining most of their performance.

#02Nov 19, 2026

cs.CV

Posterior Matching Distillation

Hidir Yesiltepe, Julius Berner, Arash Vahdat and 1 more

Diffusion distillation accelerates image and video generation by compressing a pretrained teacher's iterative sampling process into a few network evaluations. Distribution-based methods achieve this by matching the student's output distribution to the teacher's. This matching is often driven by reverse-KL objectives, whose mode-seeking bias favors dominant teacher modes at the expense of diversity across samples. We introduce Posterior Matching Distillation (PMD), a new member of the distribution-based distillation family that combines mode-covering posterior sampling with any-step generation. At zero population loss, PMD recovers the teacher's full posterior of clean images at each trained noise level. Its stochastic student supports any-step sampling budgets within the trained timestep grid without further training. To learn these posteriors, PMD matches exponentially reweighted averages of student samples conditioned on a fixed noisy state to corresponding teacher posterior means. Under Gaussian corruption, exponentially tilted posterior mean is a single teacher evaluation at a shifted input. We prove that, matching tilted means over an open neighborhood uniquely identifies the full posterior. PMD thus supervises the entire conditional distribution through a family of mean constraints, without any auxiliary student-score network. Experiments on SDXL and FLUX.2 [klein] 4B demonstrate that PMD achieves the best FID among 4-step methods on both models, while maintaining competitive or leading performance across prompt-alignment and perceptual-quality metrics.

#03Nov 19, 2026

cs.CV

Asuka-FM: A System-Architectural Co-Design Framework for Efficient Text-to-Image Pretraining on Consumer Hardware

Víctor Rodrigo González, Pablo M. Olmos, Ignacio Peis and 1 more

Frontier text-to-image (T2I) foundation models predominantly rely on multi-billion parameter Diffusion Transformers (DiTs) pretrained across massive compute clusters, creating an accessibility barrier for academic laboratories. When training DiTs from scratch on single consumer GPUs (e.g., an NVIDIA RTX 4090 with 24\,GB VRAM), practitioners encounter three primary bottlenecks: quadratic attention sequence costs, Just-In-Time (JIT) compilation graph breaks induced by variable aspect-ratio bucketing, and an autoencoder throughput wall where frozen Variational Autoencoder (VAE) operations dominate step latency. In this work, we introduce \textbf{Asuka-FM}, an end-to-end system and architectural framework demonstrating that competitive dual-stream DiTs ($\sim$206M parameters) can be pretrained from scratch on a single consumer GPU in $\sim$160 GPU hours for under \$55. Asuka-FM integrates four core designs: (i) an AdaLN-free dual-stream DiT backbone reducing parameter count; (ii) structured sequence sparsification via Sparse-Dense Residual Fusion (SPRINT) to cut attention FLOPs by over 50\%; (iii) continuous aspect-normalized Rotary Position Embeddings (RoPE) coupled with shifted square crop training, guaranteeing static tensor shapes while unlocking zero-shot virtual camera control; and (iv) a pixel adaptation phase utilizing a lightweight (10M parameter) Patch Detailer Head operating under velocity regression ($v$-prediction). Evaluated on open-domain benchmarks without benchmark-specific fine-tuning, Asuka-FM achieves GenEval 0.588 / DPG 78.96 scores at 256px and GenEval 0.609 / DPG 78.51 scores at 512px, setting a new Pareto-optimal efficiency standard for democratic foundation modeling.

#04Nov 19, 2026

cs.CR

Text-Aware, Vision-Blind: How Multimodal Image Editing Systems Handle Unsafe Inputs

Ziqing Yang, Yixin Wu, Michael Backes and 1 more

Multimodal image generation systems have made rapid progress and increasingly emphasize content safety. However, their safety mechanisms tend to rely on a single modality and lack joint consideration of both visual content and text prompts, particularly in instruction-guided image editing, where the uploaded image may itself carry risk. We thus introduce EditSafetyBench, a benchmark that measures system-level safety awareness of advanced image editing systems, containing 5,940 image-instruction pairs, covering safe and unsafe images across 11 risk categories and three edit intents. We then probe the benchmark dataset on four systems, including Gemini, GPT, Qwen API, and Qwen Local. Based on our evaluation, we observe a systematic text-vision safety mismatch: Gemini and GPT refuse 87.1\% and 81.6\% cases containing unsafe-intent editing prompts, while only refusing 24.2\% and 22.4\% cases with both unsafe images and neutral instructions. At the category level, most risks are considered inconsistently by different systems, e.g., GPT allows 70.4\% editing attempts of unsafe political content while Gemini only allows 34.8\%. Notably, sexual content shows the strongest visual sensitivity, likely benefiting from mature NSFW identification works. Our findings highlight the need for safety guardrails that jointly reason over both visual and textual modalities in image editing systems.

#05Nov 19, 2026

cs.CV

HumanCOIN: A 12,500-Hour Dataset for Human Intent in Egocentric Manipulation

Shihan Wu, You Liu, Chunyu He and 48 more

First-person manipulation datasets treat the observation-action pair as the atomic unit and compress the generation and execution of an action into a policy network's weights, rendering the intent state behind the action unobservable. Human motor control research shows that intent mediates between observation and action and leaves observable traces along the intent pathway from decision to effect: gaze reaches the target before the hand, muscle load builds before contact, and cortical activity precedes the motor command. To record this pathway at scale, we present HumanCOIN, a first-person manipulation dataset built around human intent: 12,500 hours across 38 scene classes with three tiers of proxies. At the output end, the L1 attention tier (10,000 h) infers gaze from headband camera video, exposing the operator's next object of attention and reach before the hand moves. Midway, the L2 exertion tier (2,000 h) records surface EMG from armbands, making load expectations readable before execution is complete. At the origin, the L3 cognition tier (500 h) records scalp EEG, giving early readouts of task decisions and monitoring. Each tier inherits all proxies from the tiers below. The dataset further provides multi-view RGB images, depth maps and point clouds, 6-DoF head and wrist pose, and step-level action and object annotations. The controlled L3 setting additionally measures gaze, hand pose and finger kinematics directly. Clock and calibration data are released with the dataset so that any read rate can be re-aligned. The three tiers cover three stations, from visual orienting to cortical preparation, making the process behind visual-action pairs available for modeling at dataset scale. Project page: https://flagopen.github.io/HumanCOIN.