朝花夕拾

Daily Vault

8월 13일 (목) (16건)

DonorRank: Donor Language Selection for Low-Resource Cross-Lingual Speech Recognition

arXiv:2608.11441v1 Announce Type: new Abstract: Low-resource automatic speech recognition (ASR) commonly relies on cross-lingual transfer, where models are adapted from higher-resource donor languages. However, selectin

·연구/, ·arxiv.org ↗

arxiv.org ↗

Stigma and Support in Online Sexual Violence Narratives on Reddit

arXiv:2608.11433v1 Announce Type: new Abstract: Online communities increasingly provide spaces where survivors of sexual violence can share their experiences and seek support. Although prior research has examined stigma

·연구/·arxiv.org ↗

arxiv.org ↗

Is Convergence Inevitable? Tracing Output Homogeneity Back to Base Models

arXiv:2608.11426v1 Announce Type: new Abstract: The lack of diversity in LM content is widely attributed to the alignment process, but how and where exactly in the pipeline this collapse begins is unknown. We argue that

·연구/, ·arxiv.org ↗

arxiv.org ↗

Self-Evolving Embodied Agents via Skill-Harness Evolution

arXiv:2608.11350v1 Announce Type: new Abstract: Embodied agents are increasingly built as systems around foundation models, where performance depends not only on model weights but also on the skills, context, action int

·연구/·arxiv.org ↗

arxiv.org ↗

Gloss-Free Representation Learning for Cross-Dataset Sign Spotting

arXiv:2608.11332v1 Announce Type: new Abstract: Sign-language research for resource-constrained languages is often limited by the cost of dense linguistic labels such as glosses, temporal boundaries, and sign order. Bro

·연구/·arxiv.org ↗

arxiv.org ↗

TRACE Bench: Task-driven Roleplay Agentic Checklist Evaluation

arXiv:2608.11236v1 Announce Type: new Abstract: Roleplay evaluation should do more than assign a single score: it should reveal which role requirements were tested, which failed, and which dialogue evidence supports the

·연구/·arxiv.org ↗

arxiv.org ↗

Retrofitting Recurrent Depth into a Pretrained Language Model: Installation, Extrapolation, Transfer, and Retention at Two Parameter Budgets

arXiv:2608.11233v1 Announce Type: new Abstract: A dense, pretrained language model can be retrofitted with recurrent depth and learn an iterative latent transition that persists after outcome-only annealing. Qwen2.5-0.5

·연구/·arxiv.org ↗

arxiv.org ↗

These mammal ancestors birthed live young far earlier than thought

Growth rings on a fossil bone suggest at least one mammal ancestor gave birth to live young rather than laying eggs 236 million years ago.

·연구/, ·sciencenews.org ↗

sciencenews.org ↗

Synchronizing Beliefs with Second-Order Theory-of-Mind in Human-Autonomy Teams

arXiv:2608.11229v1 Announce Type: new Abstract: Comparative feedback, asking people which of two behaviors they prefer, has become a standard way to align robot and agent behavior with human intent when the reward itsel

·연구/, ·arxiv.org ↗

arxiv.org ↗

Forecasting Side Effects of Activation Steering

arXiv:2608.11227v1 Announce Type: new Abstract: Activation steering modifies a language model by adding a learned direction to its hidden activations, enabling targeted behavioral changes without retraining. While effec

·연구/·arxiv.org ↗

arxiv.org ↗

LLMs in Process Diagram Engineering: From Optimal PFDs to Validated P&IDs

arXiv:2608.11220v1 Announce Type: new Abstract: Nowadays, the creation of a process flow diagram (PFD) and its subsequent transformation into a piping and instrumentation diagram (P&ID) is predominantly performed manual

·연구/, ·arxiv.org ↗

arxiv.org ↗

From Monolithic to Modular: Segment-level Automatic Prompt Optimization

arXiv:2608.11219v1 Announce Type: new Abstract: Automatic Prompt Optimization (APO) often rewrites prompts monolithically, which can improve one behavior while degrading others. We present SAPO, a segment-level APO meth

·연구/, ·arxiv.org ↗

arxiv.org ↗

MaSRead: Content-Addressed Reading of Replicated Latent Stores

arXiv:2608.11218v1 Announce Type: new Abstract: Independent agents that reason in latent space can share computed state as key-value cache fragments rather than text. Merged by a conflict-free replicated data type, thes

·연구/, ·arxiv.org ↗

arxiv.org ↗

AutoWorldModel-Bench: A State-Centric Benchmark for Automated World-Model Research

arXiv:2608.11216v1 Announce Type: new Abstract: World modeling is an unsettled field: architectures, training objectives, and state representations interact in complex ways, and no single recipe dominates across environ

·연구/·arxiv.org ↗

arxiv.org ↗

Detecting a Route Flip Is Easier Than Knowing Whether to Fix It: Causal Route-Mediated Damage in Quantized Mixture-of-Experts

arXiv:2608.11212v1 Announce Type: new Abstract: Top-k Mixture-of-Experts (MoE) routing is discontinuous, so a deployment-motivated numerical disturbance -- simulated 4-bit KV-cache quantization read by a protected BF16

·연구/·arxiv.org ↗

arxiv.org ↗

Distribird: Literature-Informed Prior Distribution Design for Bayesian Model Calibration

arXiv:2608.11210v1 Announce Type: new Abstract: Bayesian calibration of process-based models requires a prior distribution for each model parameter. Despite decades of methodological work, researchers almost always fall

·연구/·arxiv.org ↗

arxiv.org ↗

8월 12일 (수) (4건)

Toddler's tragic death from brain-destroying amoeba offers lessons for doctors

Boy's unusual symptoms could help identify the next case sooner.

·연구/, ·arstechnica.com ↗

arstechnica.com ↗

Have physicists finally discovered glueballs? New evidence points to yes.

“It’s the strongest evidence yet that particles dominated by a glueball component can exist in nature.”

·연구/·arstechnica.com ↗

arstechnica.com ↗

MIT selected to lead new NSF materials research center

The Materials Research Science and Engineering Center unites researchers across disciplines to develop technologies for medical imaging, sustainable metals production, and next-generation electronics.

·연구/, ·news.mit.edu ↗

news.mit.edu ↗

Estrogen therapy is linked to fewer signs of Alzheimer’s in the brain

Estrogen-only menopause therapy in older women was linked to less Alzheimer’s brain pathology at autopsy but may not prevent the disease.

·연구/·sciencenews.org ↗

sciencenews.org ↗

아침 이슬이 마르기 전에, 꽃은 이미 길을 떠났어요.