In reasoning chains generated by large language models (LLMs), initial errors often propagate and undermine the reliability of the final conclusion.Current LLM-based error detection methods often fail to detect propagated errors because earlier errors can corrupt judgments of downstream reasoning.To better detect such errors, we introduce Autoregressive Reasoning Entailment Stability (ARES), a probabilistic framework that evaluates each reasoning step based solely on previously-verified premises.This inductive method yields a nuanced score for each step and provides certified statistical guarantees of its soundness, rather than a brittle binary label.ARES achieves state-of-the-art performance across four benchmarks (72.1% Macro-F1, +8.2 points) and demonstrates superior robustness on very long synthetic reasoning chains, where it excels at detecting propagated errors (90.3% F1, +27.6 points). 1

We study how to subvert large language models (LLMs) from following prompt-specified rules. We first formalize rule-following as inference in propositional Horn logic, a mathematical system in which rules have the form "if $P$ and $Q$, then $R$" for some propositions $P$, $Q$, and $R$. Next, we prove that although small transformers can faithfully follow such rules, maliciously crafted prompts can still mislead both theoretical constructions and models learned from data. Furthermore, we demonstrate that popular attack algorithms on LLMs find adversarial prompts and induce attention patterns that align with our theory. Our novel logic-based framework provides a foundation for studying LLMs in rule-based settings, enabling a formal analysis of tasks like logical reasoning and jailbreak attacks.

Layer-wise preconditioning methods are a family of memory-efficient optimization algorithms that introduce preconditioners per axis of each layer's weight tensors. These methods have seen a recent resurgence, demonstrating impressive performance relative to entry-wise ("diagonal") preconditioning methods such as Adam(W) on a wide range of neural network optimization tasks. Complementary to their practical performance, we demonstrate that layer-wise preconditioning methods are provably necessary from a statistical perspective. To showcase this, we consider two prototypical models, *linear representation learning* and *single-index learning*, which are widely used to study how typical algorithms efficiently learn useful *features* to enable generalization. In these problems, we show SGD is a suboptimal feature learner when extending beyond ideal isotropic inputs $\mathbf{x} \sim \mathsf{N}(\mathbf{0}, \mathbf{I})$ and well-conditioned settings typically assumed in prior work. We demonstrate theoretically and numerically that this suboptimality is fundamental, and that layer-wise preconditioning emerges naturally as the solution. We further show that standard tools like Adam preconditioning and batch-norm only mildly mitigate these issues, supporting the unique benefits of layer-wise preconditioning.

Stability guarantees have emerged as a principled way to evaluate feature attributions, but existing certification methods rely on heavily smoothed classifiers and often produce conservative guarantees. To address these limitations, we introduce soft stability and propose a simple, model-agnostic, sample-efficient stability certification algorithm (SCA) that yields non-trivial and interpretable guarantees for any attribution method. Moreover, we show that mild smoothing achieves a more favorable trade-off between accuracy and stability, avoiding the aggressive compromises made in prior certification methods. To explain this behavior, we use Boolean function analysis to derive a novel characterization of stability under smoothing. We evaluate SCA on vision and language tasks and demonstrate the effectiveness of soft stability in measuring the robustness of explanation methods.

Anomaly detection is widely used for identifying critical errors and suspicious behaviors, but current methods lack interpretability. We leverage common properties of existing methods and recent advances in generative models to introduce counterfactual explanations for anomaly detection. Given an input, we generate its counterfactual as a diffusion-based repair that shows what a non-anomalous version $\textit{should have looked like}$. A key advantage of this approach is that it enables a domain-independent formal specification of explainability desiderata, offering a unified framework for generating and evaluating explanations. We demonstrate the effectiveness of our anomaly explainability framework, AR-Pro, on vision (MVTec, VisA) and time-series (SWaT, WADI, HAI) anomaly datasets. The code used for the experiments is accessible at: https://github.com/xjiae/arpro.

Explanation methods for machine learning models tend not to provide any formal guarantees and may not reflect the underlying decision-making process. In this work, we analyze stability as a property for reliable feature attribution methods. We prove that relaxed variants of stability are guaranteed if the model is sufficiently Lipschitz with respect to the masking of features. We develop a smoothing method called Multiplicative Smoothing (MuS) to achieve such a model. We show that MuS overcomes the theoretical limitations of standard smoothing techniques and can be integrated with any classifier and feature attribution method. We evaluate MuS on vision and language models with various feature attribution methods, such as LIME and SHAP, and demonstrate that MuS endows feature attributions with non-trivial stability guarantees.

We present a type-theoretic framework for data stream processing for real-time decision making, where the desired computation involves a mix of sequential computation, such as smoothing and detection of peaks and surges, and naturally parallel computation, such as relational operations, key-based partitioning, and map-reduce. Our framework unifies sequential (ordered) and relational (unordered) data models. In particular, we define synchronization schemas as types, and series-parallel streams (SPS) as objects of these types. A synchronization schema imposes a hierarchical structure over relational types that succinctly captures ordering and synchronization requirements among different kinds of data items. Series-parallel streams naturally model objects such as relations, sequences, sequences of relations, sets of streams indexed by key values, time-based and event-based windows, and more complex structures obtained by nesting of these. We introduce series-parallel stream transformers (SPST) as a domain-specific language for modular specification of deterministic transformations over such streams. SPSTs provably specify only monotonic transformations allowing streamability, have a modular structure that can be exploited for correct parallel implementation, and are composable allowing specification of complex queries as a pipeline of transformations.

We present counterfactual symbolic execution, a new approach that produces counterexamples that localize the causes of failure of static verification. First, we develop a notion of symbolic weak head normal form and use it to define lazy symbolic execution reduction rules for non-strict languages like Haskell. Second, we introduce counterfactual branching, a new method to identify places where verification fails due to imprecise specifications (as opposed to incorrect code). Third, we show how to use counterfactual symbolic execution to localize refinement type errors, by translating refinement types into assertions. We implement our approach in a new Haskell symbolic execution engine, G2, and evaluate it on a corpus of 7550 errors gathered from users of the LiquidHaskell refinement type system. We show that for 97.7% of these errors, G2 is able to quickly find counterexamples that show how the code or specifications must be fixed to enable verification.