Research Papers 论文研究 7h ago Updated 2h ago 更新于 2小时前 45

Sparse Readout Prism: Explaining Logit-Lens Scores in Features Instead of Tokens 稀疏读出棱镜:用特征而非词元解释Logit-Lens分数

The paper introduces Sparse Readout Prism (SRP), a method that decomposes the readout (unembedding) matrix of language models into sparse features, enabling analysis independent of any fitting corpus The authors demonstrate "corpus conditionality": two lenses fit on different corpora can assign different tokens to the same hidden states, revealing a fundamental limitation of existing lens methods SRP expresses any token logit or logit difference as a sum of contributions from sparse readout feat 提出Sparse Readout Prism (SRP)方法,将logit-lens得分从token层面分解到稀疏readout特征层面 揭示lens方法的"语料库条件依赖性"问题:不同拟合语料库的lens对相同隐藏状态会输出不同token SRP仅基于权重分解readout,无需语料库,提供独立于拟合语料库的分析控制 实验表明SRP比六种几何基线方法多重建8.9-17.3个百分点的logit差 主导readout特征具有跨语料库稳定性,为lens分析提供新的可解释性单元

58
Hot 热度
72
Quality 质量
65
Impact 影响力

Analysis 深度分析

TL;DR

  • The paper introduces Sparse Readout Prism (SRP), a method that decomposes the readout (unembedding) matrix of language models into sparse features, enabling analysis independent of any fitting corpus
  • The authors demonstrate "corpus conditionality": two lenses fit on different corpora can assign different tokens to the same hidden states, revealing a fundamental limitation of existing lens methods
  • SRP expresses any token logit or logit difference as a sum of contributions from sparse readout features, establishing features as a new unit of analysis for interpretability
  • SRP's sparse approximation reconstructs 8.9–17.3 percentage points more of tested logit differences than the strongest of six geometric baselines
  • While token-level readings vary with fitting corpus, the dominant readout features remain stable, providing a corpus-independent control for mechanistic interpretability research

Why It Matters

This work addresses a critical confound in mechanistic interpretability: logit-lens readings conflate hidden state representations with the readout matrix, making it difficult to determine whether observed patterns reflect genuine internal representations or artifacts of the decoding process. By isolating readout structure from corpus-dependent fitting, SRP provides researchers with a more reliable tool for comparing interpretations across models, layers, and training conditions.

Technical Details

  • Sparse Readout Prism (SRP) decomposes the unembedding matrix using only its weights (no corpus required), expressing token logits and logit differences as additive contributions from sparse readout features
  • The paper identifies and formalizes corpus conditionality — the phenomenon where lenses fitted on different corpora produce divergent token assignments for identical hidden states, undermining cross-study comparability
  • Evaluation: SRP outperforms six geometric baselines (based on readout row relationships) by 8.9–17.3 percentage points in logit difference reconstruction accuracy
  • Ablation studies confirm that shifting SRP features shifts logit differences proportionally to their computed contributions, validating the decomposition's causal interpretability
  • The dominant readout features prove stable across corpora, unlike token-level readings, making SRP a corpus-independent control for lens-based analyses

Industry Insight

  • Interpretability tooling should move beyond token-level explanations; feature-level decompositions like SRP offer more stable and comparable insights across model variants and training regimes
  • Researchers using logit-lens or related methods should account for corpus conditionality when comparing results across studies, as differing fitting corpora may produce misleading apparent discrepancies
  • SRP's corpus-free construction makes it particularly valuable for benchmarking and auditing models where training data is proprietary or unavailable, enabling readout analysis without data access

TL;DR

  • 提出Sparse Readout Prism (SRP)方法,将logit-lens得分从token层面分解到稀疏readout特征层面
  • 揭示lens方法的"语料库条件依赖性"问题:不同拟合语料库的lens对相同隐藏状态会输出不同token
  • SRP仅基于权重分解readout,无需语料库,提供独立于拟合语料库的分析控制
  • 实验表明SRP比六种几何基线方法多重建8.9-17.3个百分点的logit差
  • 主导readout特征具有跨语料库稳定性,为lens分析提供新的可解释性单元

为什么值得看

这篇论文揭示了当前logit-lens方法的一个根本性缺陷——结果受拟合语料库影响,并提出了不依赖语料库的SRP方法作为解决方案。对于从事模型可解释性研究的AI从业者来说,这提供了更可靠的lens分析框架,有助于深入理解语言模型内部表示。

技术解析

  • SRP方法仅使用unembedding矩阵的权重进行分解,将token logit或logit差表示为稀疏readout特征的贡献之和,完全独立于任何训练语料库
  • 论文提出"语料库条件依赖性"概念,证明两个仅在拟合语料库上不同的lens可以为相同的隐藏状态报告不同的token
  • 实验评估显示,用SRP稀疏近似替换原始readout,在重建测试logit差方面比六种基于readout行几何关系的基线方法高出8.9-17.3个百分点
  • 特征消融实验表明,移除特定readout特征会按其SRP贡献比例改变logit差,验证了特征贡献的可解释性
  • 尽管token层面的lens读取随拟合语料库变化,但主导readout特征保持稳定,证明其特征层面的分析更具鲁棒性

行业启示

  • 可解释性研究需要警惕方法本身的拟合偏差,SRP提供的语料库无关控制为lens分析树立了新标准
  • 从token级分析转向特征级分析可能揭示被token身份遮蔽的模型内部结构,为机制可解释性开辟新路径
  • 建议研究者在比较不同lens方法或跨模型分析时,采用SRP等不依赖语料库的方法以确保结果的可比性

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Research 科学研究 LLM 大模型 Evaluation 评测