KOINEU

SOM Directions are Better than One: Multi-Directional Refusal Suppression in Language Models

1. 연구 배경 및 동기 거부 행동 은 현재 LLM 안전성 연구의 핵심 과제이며, “거부를 억제”하는 jailbreak 공격에 대한 방어 메커니즘을 이해하는 것이 필수적이다. 기존 메커니즘 해석 연구는 단일 벡터 (difference‑in‑means)로 개념을 요약했지만, 이는 다중 차원적 인 개념 구조를 과도하게 단순화한다는 비판을 받아왔다. Self‑Organizing Map 은 고차원 데이터를 저차원 격자 형태의 토폴로지 보존 맵으로 변환해, 데이터 군집을 시각화하고 다중 프로토타입 을 자연스럽게 추출한다는 장점이 있다. 2.

SOM Directions are Better than One: Multi-Directional Refusal Suppression in Language Models

Spatio-temporal air flow properties in a 3D personalised model of the human lung

Spilling the Beans: Teaching LLMs to Self-Report Their Hidden Objectives

Standards and Safety: an Overview

Stimulated interactions of low-energy free-electrons with light

Streaming Tensor Programs: A Streaming Abstraction for Dynamic Parallelism

Structural barriers of the discrete Hasimoto map applied to protein backbone geometry

Symbolic Neural Generation with Applications to Lead Discovery in Drug Design

Systematic Evaluation of Single-Cell Foundation Model Interpretability Reveals Attention Captures Co-Expression Rather Than Unique Regulatory Signal

Temperature and Respiratory Emergency Department Visits: A Mediation Analysis with Ambient Ozone Exposure

The Cascade Equivalence Hypothesis: When Do Speech LLMs Behave Like ASR$rightarrow$LLM Pipelines?

The Hidden Nature of Non-Markovianity

The Impact of Formations on Football Matches Using Double Machine Learning. Is it worth parking the bus?

The Limits of Obliviate: Evaluating Unlearning in LLMs via Stimulus-Knowledge Entanglement-Behavior Framework

The Stark effect in molecular Rydberg states: Calculation of Rydberg-Stark manifolds of H$_2$ and D$_2$ including fine and hyperfine structures

The Tool Decathlon: Benchmarking Language Agents for Diverse, Realistic, and Long-Horizon Task Execution

Tolerances to driver-witness misalignment in a quasilinear plasma wakefield accelerator

Tongyi DeepResearch Technical Report

Towards Scalable Meta-Learning of near-optimal Interpretable Models via Synthetic Model Generations

Tracking the Brownian motion of DNA-functionalized magnetic nanoparticles for conformation analysis beyond the optical resolution limit

Transition from traveling fronts to diffusion-limited growth in expanding populations

Tunable asymmetric swimming in biflagellate microswimmers

U-FedTomAtt: Ultra-lightweight Federated Learning with Attention for Tomato Disease Recognition

UCO: A Multi-Turn Interactive Reinforcement Learning Method for Adaptive Teaching with Large Language Models

Uncovering Bugs in Formal Explainers: A Case Study with PyXAI

UniField: Joint Multi-Domain Training for Universal Surface Pressure Modeling

UrbanVLA: A Vision-Language-Action Model for Urban Micromobility

Vibrational infrared and Raman spectra of the methanol molecule with equivariant neural-network property surfaces

Vision-Language-Policy Model for Dynamic Robot Task Planning

VitalBench: A Rigorous Multi-Center Benchmark for Long-Term Vital Sign Prediction in Intraoperative Care

VocalNet-M2: Advancing Low-Latency Spoken Language Modeling via Integrated Multi-Codebook Tokenization and Multi-Token Prediction

What We Don't C: Representations for scientific discovery beyond VAEs

Who Sees the Risk? Stakeholder Conflicts and Explanatory Policies in LLM-based Risk Assessment

Whodunnit? The case of midge swarms

Wide-Surface Furnace for In Situ X-Ray Diffraction of Combinatorial Samples using a High-Throughput Approach

'I Like That You Have to Poke Around': Instructors on How Experiential Approaches to AI Literacy Spark Inquiry and Critical Thinking

A fluctuating lattice Boltzmann formulation based on orthogonal central moments

A Pragmatic View of AI Personhood

A Unified Formulation for $langle hat{S}^2 rangle $ in Two-Component TDDFT

Accelerated, Memory-Efficient Far-Field Scattering Computation with Monte Carlo SBR

Activation-Space Uncertainty Quantification for Pretrained Networks

Adaptive Data Flywheel: Applying MAPE Control Loops to AI Agent Improvement

AI Code in the Wild: Measuring Security Risks and Ecosystem Shifts of AI-Generated Code in Modern Software

AIOT based Smart Education System: A Dual Layer Authentication and Context-Aware Tutoring Framework for Learning Environments

An Agentic Framework for Rapid Deployment of Edge AI Solutions in Industry 5.0

An Electrocardiogram Multi-task Benchmark with Comprehensive Evaluations and Insightful Findings

An Indoor Radio Mapping Dataset Combining 3D Point Clouds and RSSI

AURA: A Reinforcement Learning Framework for AI-Driven Adaptive Conversational Surveys

Automated Histopathology Report Generation via Pyramidal Feature Extraction and the UNI Foundation Model

AVOID-JACK: Avoidance of Jackknifing for Swarms of Long Heavy Articulated Vehicles

< Category Statistics (Total: 5060) >

Start searching

No results found