This paper develops a categorical framework -- Learning in Infinitesimal Non-Compositional Sketches (LINCS) -- as the repair of non-compositionality: failures of diagrams to factor through quotient sketches lifted to the tangent category setting. Machine learning problems are specified as sketches: graphs with commutativity conditions $\mathcal D$, limit cones $\mathcal L$, and colimit cocones $\mathcal K$, generalizing the usual scalarization of loss functions or vector space assumptions. Non-compositionality is defined purely as failure of a universal factorization problem, not as arithmetic error between the desired and actual predictions. Given a learning sketch $\mathbb S=(S,\mathcal D,\mathcal L,\mathcal K)$, whose underlying graph is $S$, and a model $D:J \rightarrow C$, the base defect is the obstruction to factorization $\mbox{Obs}(\mbox{Fact}_{\mathbb S}(D))$. The tangent lift applies the tangent functor $T$ to obtain $TD:J \rightarrow C$, and LINCS is defined as the obstruction $\mbox{Obs}(\mbox{Fact}_{\mathbb S}(TD))$ -- asking whether infinitesimal perturbations preserve the compositionality constraints.The paper also introduces Tangent Learning Sketches, which are sketches equipped with Cockett-Cruttwell tangent structure. The paper defines the INC endofunctor, which iterates the tangent lift, producing a tower $D,TD,T^2D, \cdots$ of factorization problems. ML is thereby formulated as the search for a coalgebraic fixed point where successive tangent unfoldings stabilize ($\nu T_{\mbox{INC}}$). Using the Aczel--Mendler theorem, we prove existence of a final INC coalgebra whenever $T_{\mbox{INC}}$ admits a set-based class realization that creates its final carrier. A detailed experimental evaluation of LINCS is underway in a number of concrete ML settings, including deep learning, large language models, and reinforcement learning, and is described in companion papers.
This work refine the existing parameter estimation guarantees under the fatness assumption, improving the prior sample complexity to $O( \log n / \epsilon^2)$ for $\ell_\infty$-recovery, matching the untruncated minimax rate.
Rohan Chauhan, Ioannis Panageas· arXiv.org· 0 citations
We prove that fully independent SparseStack achieves the oblivious subspace embedding parameters conjectured by Nelson and Nguyen (FOCS 2013): $m=O((d+\log(1/\delta))/\varepsilon^2)$ rows and $s=O(\log(d/\delta)/\varepsilon)$ nonzero entries per column for distortion $\varepsilon$ and failure probability $\delta$ on any fixed $d$-dimensional subspace, with explicit constants. The proof turns random-matrix concentration into a problem in finite-dimensional linear algebra. A coupling first reduces the moment estimates to a model with independent finite-valued entries. We represent these variables by multiplication operators, so their matrix moments become exact matrix elements of a deterministic operator on a finite tensor product. The central estimate bounds the contribution of $\ell\ge1$ occupied sites sharing the external factor $\mathbb{R}^d$ by $d+\ell-1$ rather than $d\ell$, yielding additive dependence on the dimension and the moment order. This approach controls both spectral edges without Gaussian comparison. The main theorem has been formally verified in Lean 4.
We introduce \textsc{LatentGamma}, a differentiable operator on the unit cube $[0,1]^M$ designed as a smooth surrogate of Tarski's immediate consequence operator $\TP$ associated with a definite Horn program $P$ on $M$ atoms. The operator is built as a five-stage composition combining sigmoidal gating, softmax routing, and a residual update that enforces monotonicity by construction. We establish four theoretical results. First, the iterated sequence is coordinate-wise non-decreasing and bounded by $\ind$, hence converges to a fixed point of \textsc{LatentGamma}; the operator itself is lattice-monotone on $[0,1]^M$. Second, our \emph{constructive expressivity theorem} shows that for every definite Horn program $P$ there exists a closed-form parameter assignment $\theta^*(P)$ and an explicit time bound $T_{\max}(P)$ such that the iterated sequence \emph{exactly} reproduces $\TPinf(F_0)$ for every initial fact set $F_0$ throughout the window $[D(P, F_0), T_{\max}(P)]$, where $D(P, F_0) \leq M$ is the derivation depth. The window $T_{\max}$ is large in practice ($\geq 10^5$ for sparse programs) and reflects the finite-time nature of computation by smooth sigmoidal gates. Third, the oracle is robust to Gaussian noise on its body and head logits, with explicit non-asymptotic bounds. Fourth, we prove a matching information-theoretic lower bound on the parameter count. We provide a complete numerical validation on programs ranging from $M = 33$ to $M = 504$ atoms: oracle accuracy reaches $1.0000$ on $2000$ test cases with zero false-positive and false-negative rates, and the empirical noise tolerance $\sigma_{\max}$ scales precisely as the union bound $RM \cdot \Phi(-10/\sigma_\eta)$ predicts.
We identify a recurrent algebraic regularity in Transformer attention: a sparse subset of effective OV operators $T=OV^\top$ nearly closes under composition, $T^2\approx\alpha T$. Across six pretrained endpoints spanning 2.8B--235B parameters, 3.98--8.00% of heads reach squared closure alignment $\mathcal{P}\geq0.9$, while no matched within-layer O/V mismatch does. An exact principal-coordinate factorization, $T=Q_OKQ_V^\top$ and $T^2=Q_O(KDK)Q_V^\top$, separates within-support transport from read--write return geometry. Across all 7,304 heads in nine MHA/GQA models, scrambling only the orientation of $K$ while preserving singular values, norms, factor spans, and principal angles reduces median closure from 0.336 to $1.04\times10^{-4}$; trained orientation wins for 98.64% of heads and in every layer. Constructive searches show that high closure is feasible in every surveyed layer, but usually not attained. Retrospective trajectories in three independently trained lineages further separate broadly available capacity from the orientations attained by final strong heads. Under exact value sharing, headwise closure extends to a right-action algebra, $T_iT_j=\alpha_jT_i$. Seven-model experiments verify the approximate law and reveal distinct oblique projections with a shared value-defined kernel. These results characterize scaled idempotence as a sparse trained orientation within broadly available geometric capacity and show how value sharing extends a headwise relation into a local operator algebra.
The same geometric analysis covers LoRA, ReZero, and zero-init adapter constructions, establishing gate-zero growth as the canonical instance of a shared local geometry that governs safe capacity activation in CL.
Comparison with the running mean highlights the stationary insertion-length distribution, its time-homogeneous update, stationary coefficient profile, and fixed effective memory, and its time-homogeneous update, stationary coefficient profile, and fixed effective memory.
Pedro M. M. de Castro· 1 citation· ⚡1
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
