arXiv research

New analysis shows interpretability doesn't guarantee steering utility in LLMs.

Optimizes hedge ratio for delta-neutral liquidity positions in AMMs.

LLMs show surprising confidence in their answers, beyond just tokens.

Entropy-based decoding improves DLM sampling efficiency.

Proposes an alternative method for quantifying uncertainty in complex models.

Enhanced Sampling Scheme improves masked generative modeling.

InfoSFT improves LLMs by focusing on informative, medium-confidence tokens.

Derives token price process for AMM tokens, finds leverage effect and pricing discrepancies.

Proposes integrating global and local entropy for more reliable LLMs.

In recent years there has been an increasing interest in learning Bayesian networks from data. One of the most effective methods for learning such networks is based on the minimum description length (MDL) principle. Previous work has shown that this learning procedure is asymptotically successful: with probability one,…

GPT learns a causal world model from token predictions, validated in game sequences.

Disagreement between two classifiers regarding the class membership of an observation in pattern recognition can be indicative of an anomaly and its nuance. As in general classifiers base their decision on class aposteriori probabilities, the most natural approach to detecting classifier incongruence is to use divergen…

Improved deep hedging with ensemble uncertainty quantification.

New method for estimating spatial associations with discrete data, even under model misspecification.

A/B testing refers to the task of determining the best option among two alternatives that yield random outcomes. We provide distribution-dependent lower bounds for the performance of A/B testing that improve over the results currently available both in the fixed-confidence (or delta-PAC) and fixed-budget settings. When…

We consider the problem of near-optimal arm identification in the fixed confidence setting of the infinitely armed bandit problem when nothing is known about the arm reservoir distribution. We (1) introduce a PAC-like framework within which to derive and cast results; (2) derive a sample complexity lower bound for near…

New methods for delta-moves on algebraically split links identified.

Delta-unlinking number measures how to unlink algebraically split links.

We develop techniques for studying fundamental groups and integral singular homology of symmetric Delta-complexes, and apply these techniques to study moduli spaces of stable tropical curves of unit volume, with and without marked points. As one application, we show that Delta_g and Delta_{g,n} are simply connected, fo…

This study examines representation bias in open-source Qwen models for investment decisions.

The paper calculates delta invariants for specific geometric structures.

TWM doesn't reduce delta in PDLPs, proving impossibility.

A Delta-groupoid is an algebraic structure which axiomitizes the combinatorics of a truncated tetrahedron. It is shown that there are relations of Delta-groupoids to rings, group pairs, and (ideal) triangulations of three-manifolds. In particular, one can associate a Delta-groupoid to ideal triangulations of knot compl…

We refine the analysis of hedging strategies for options under the SABR model carried out in [2]. In particular, we provide a theoretical justification of the empirical observation made in [2] that the modified delta ("Bartlett's delta") introduced there provides a more accurate and robust hedging strategy than the con…

A Delta-groupoid is an algebraic structure which axiomatizes the combinatorics of a truncated tetrahedron. By considering two simplest examples coming from knot theory, we illustrate how can one associate a Delta-groupoid to an ideal triangulation of a three-manifold. We also describe in detail the rings associated wit…

Link-homotopy and self Delta-equivalence are equivalence relations on links. It was shown by J. Milnor (resp. the last author) that Milnor invariants determine whether or not a link is link-homotopic (resp. self Delta-equivalent) to a trivial link. We study link-homotopy and self Delta-equivalence on a certain componen…

Continuity of delta invariant leads to uniform Kähler-Einstein metrics.

Bootstrap method for Markov chains in reinforcement learning.

Emergent misalignment is influenced by training dynamics, model priors, and data.

In this paper we construct some invariants of spatial graphs by disk-summing the constituent knots and show the delta edge-homotopy invariance of them. As an application, we show that there exist infinitely many slice spatial embeddings of a planar graph up to delta edge-homotopy, and there exist infinitely many bounda…

Delta method vs Bootstrap for deep learning classification shows strong linear relationship and faster computation.

Study delta invariant of curves on rational surfaces using topological methods.

Study shows singular set of distance functions is delta-convex.

Delta finite-type invariants are defined analogously to finite-type invariants, using delta moves instead of crossing changes. We show that they are closely related to the lower central series of the commutator subgroup of the pure braid group.

We call a Delta Diagram any diagram of a knot or link whose regions (including the unbounded one) have 3, 4, or 5 sides. We prove that any knot or link admits a delta diagram. We define and estimate combinatorial link invariants stemming from this definition.

The paper accelerates LLM inference by adding early exit heads trained in a self-supervised manner.

Neural sequence generation is typically performed token-by-token and left-to-right. Whenever a token is generated only previously produced tokens are taken into consideration. In contrast, for problems such as sequence classification, bidirectional attention, which takes both past and future tokens into consideration, …

Study virtualized Delta, Sharp, and Pass moves for oriented virtual knots and links.

The paper classifies pretzel links with 2 components and gives conditions for those with 3 or more.

We study generalizations of finite-type knot invariants obtained by replacing the crossing change in the Vassiliev skein relation by some other local move, analyzing in detail the band-pass and doubled-delta moves. Using braid-theoretic techniques, we show that, for a large class of local moves, generalized Goussarov's…

Lower bounds for delta invariant of weighted hypersurfaces proved for K-stability.

Study calculates liquidity costs for delta hedging of European options.

This paper uses Malliavin calculus to price and compute delta of financial derivatives in jump-diffusion models.

Delta Variances efficiently estimate epistemic uncertainty in neural networks.

We study the calibration of several state of the art neural machine translation(NMT) systems built on attention-based encoder-decoder models. For structured outputs like in NMT, calibration is important not just for reliable confidence with predictions, but also for proper functioning of beam-search inference. We show …

We prove that the crossing changes, Delta moves, and sharp moves are unknotting operations on welded knots.

Paper reduces vocabulary losslessly for language model cooperation.

This research improves capital efficiency and impermanent loss in cryptocurrency markets using multi-token trading pools.