cs.CLNov 14, 2026

The Geometry of Conceptors in Neural-Activation Spaces

Authors: Ilias Triantafyllopoulos, Young Min Cho, Ren Tao, Miranda Muqing Miao, Sunny Rai, Lyle Ungar, Sharath Chandra Guntuku, Sunghye Cho, 

Organizations: New York University · University of Pennsylvania · University of Cincinnati · Linguistic Data Consortium, University of Pennsylvania

Abstract

Conceptors represent concepts as graded collections of directions in neural-activation space rather than as single vectors. We develop a geometric account of how these representations depend on the contrast used to define a concept: opposing values in bipolar fitting, or presence and absence in unipolar fitting. We derive an exact decomposition of a centered pooled conceptor into an average pole filter, a mean-separation update, and a covariance-shape update. Experiments across five language models on sentiment, political leaning and stance, and depression expression show that both updates improve reconstruction of held-out contrasts. Additional directions preserve variation beyond the mean-difference line, and changing the comparator changes the benefit of these directions. Shape, size, and overlap each show up downstream: in amplification experiments, bipolar fits raise evidence for both poles more consistently than unipolar fits; conceptor size (the quota) correlates positively with layerwise probe performance; and cross-concept overlap quantifies interference under subspace removal, linking shared directions to Boolean composition. In an additional generation evaluation, conceptor filters achieve higher target-expression win rates against an unsteered reference than three mean-based controls. Together, these findings explain how fitting choices shape conceptor geometry and support bipolar fitting when the objective is to amplify both opposing poles.

Figures & tables

Appendix figures & tables36 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

May 6, 2026

Conceptors for Semantic Steering

Activation-based steering provides control of LLM behavior at inference time, but the dominant paradigm reduces each concept to a single direction whose geometry is left largely unexamined. Rather than selecting a single steering direction, we use conceptors: soft projection matrices estimated from activations pooled across both poles of a bipolar concept, which preserve the concept's full multidimensional subspace. A geometric analysis shows the bipolar subspace strictly subsumes the single-vector baseline. We further show that the conceptor quota provides a parameter-free layer-selection diagnostic, predicting concept separability with Pearson correlations up to r=0.96 across three instruction-tuned models and three semantic dimensions. Beyond selection, conceptors admit a closed-form Boolean algebra (AND, OR, NOT): we evaluate conceptor compositionality on thematically related sub-concepts. Across a systematic five-axis design-space evaluation, conceptors match or outperform additive baselines at layers where concept subspaces are multi-dimensional while producing substantially fewer degenerate outputs. Conceptor steering is a geometrically principled, compositional, and practically safer alternative to single-direction steering from a limited number of contrastive pairs.
Sep 30, 2026

Concept Subspaces Compute Beyond the Logit Lens: A Weights-Only Test for Locating Representations Upstream of Readout

A concept subspace's effect on model behavior does not establish how it relates to the output readout. We introduce a two-sided geometric diagnostic that measures an extracted subspace's overlap with the dominant right-singular directions of the unembedding matrix, evaluated against output-oriented positive controls. Given an extracted basis, the raw diagnostic requires only model weights. Our testbed is the Format-Agnostic Reasoning Subspace (FARS), a ten-dimensional basis extracted from eighteen reasoning concepts expressed in six surface forms. Across nine rank-matched estimators and twenty-six models, four activation-derived concept estimators carry only 0.38--0.80% mean energy in the top-ten readout span. Final-layer PCA carries 3.56%, exceeding FARS in 25 of 26 models. A same-layer next-token control, evaluated using a fitted linear translator for depth matching, carries approximately thirteen times more energy than FARS, with separation in all 25 tested models. Re-extracting FARS on ten disjoint concepts yields 62--100% cross-format retrieval across twenty-four generative models, demonstrating transfer of the extraction procedure rather than a fixed basis. A complementary four-model, three-seed intervention study finds model-dependent source-directed effects that remain well below full-vector replacement. Together, the geometry and intervention controls distinguish concept structure from dominant readout directions while limiting claims of causal sufficiency.
Dec 4, 2025

The SuperActivator Mechanism: Transformers Concentrate Reliable Concept Signals in the Tail

Concept vectors aim to enhance model interpretability by linking learned internal representations with human-understandable semantics, but their practical utility is often limited by noisy and inconsistent activations. In this work, we discover the SuperActivator Mechanism: a transformer dynamic that amplifies concept activation gaps, concentrating the most reliable concept evidence into a small set of highly activated tokens, which we call SuperActivators. We theoretically characterize this mechanism, proving that concept-aligned attention produces two competing effects: a tail amplification effect, where gaps among already-high activations grow disproportionately fast, and an equalization effect that eventually dominates as attention concentrates on extreme tokens. We validate the predicted tail amplification empirically, finding that in-concept activation distributions develop well-separated upper tails consistently across models, image and text modalities, and concept types. Moreover, we demonstrate the practical implications of SuperActivators for concept detection and localization. First, using only high-activation tokens as concept indicators improves detection by up to 14 F1 points over competing baselines. Second, restricting detection to the extreme activation tail is not merely sufficient but essential: including activations beyond the top 5-10% of the in-concept distribution introduces noise that degrades performance. Finally, using SuperActivators as instance-specific attribution targets yields more accurate and faithful concept localization heatmaps than using global concept vectors.