Research
On-device research index

arXiv research

A locally-built, LLM-digested index of recent arXiv papers in quant finance, geometry/topology, and statistical ML — keyword search served straight from SQLite on this machine.

168,695 papers · 148 categories

Trend · papers per month

52105157209 · Jun 202019922001200920172026
48 results for looped transformers

Training-free looped transformers improve model performance without additional training.

problem Improving model performance without additional training or fine-tuning.
method A lightweight inference-time wrapper loops a contiguous mid-stack block of layers of a frozen checkpoint without additional fine-tuning.
result Our method improves model performance across various model families.

Looped transformers outperform standard transformers in complex reasoning tasks due to a specific loss landscape geometry.

problem Understanding why looped transformers outperform standard transformers in complex reasoning tasks.
method Explained through loss landscape geometry, distinguishing between U-shaped and V-shaped valleys, and proposing SHIFT training strategy.
result Looped transformers' recursive architecture induces a River-V-Valley landscape, leading to better loss convergence and complex pattern learning.

Study fundamental groups of geometric transformation groups using loop spaces.

problem Understanding fundamental groups of geometric transformation groups.
method Use differential forms on loop spaces to prove infinite fundamental groups.
result Proves infinite fundamental groups for specific geometric transformation groups.

Looped transformers with LN converge to power method for principal component prediction.

problem Understanding how transformers learn algorithmic procedures.
method Study of principal component prediction with looped linear transformers and layer normalization.
result Gradient descent trains looped transformers with LN to implement the power method for principal component prediction.

Looped Transformers improve robustness and expressivity in in-context learning for diverse tasks.

problem Improving robustness and expressivity in in-context learning for diverse tasks.
method Study in-context linear regression with diverse tasks, focusing on depth and looping.
result Looped Transformers exhibit similar expressive power and are provably robust under mild assumptions.

We introduce a property of mutation loops, called the sign stability, with a focus on an asymptotic behavior of the iteration of the tropical X\mathcal{X}-transformation. A sign-stable mutation loop has a numerical invariant which we call the cluster stretch factor, in analogy with that of a pseudo-Anosov mapping clas…

2019-11-18abs ↗pdf ↗

Researchers study injectivity of magnetic and thermostatic nonabelian ray transforms on compact surfaces.

problem Injectivity of magnetic and thermostatic nonabelian ray transforms on compact surfaces.
method Loop group factorization method for nontrapping λλ-geodesic flows and the general linear group of invertible complex matrices.
result General injectivity question of the nonabelian ray transform for simple magnetic flows is settled.

The paper provides a link between ergodic theory and symplectic topology. A classical notion of ergodic theory is a skew product map associated with a loop in a group of transformations. We study skew products which come from loops in the group of Hamiltonian diffeomorphisms of a symplectic manifold. Our main question …

1998-06-29abs ↗pdf ↗

The loop space of the Riemann sphere consisting of all CkC^k or Sobolev Wk,pW^{k,p} maps from the circle S1S^1 to the sphere is an infinite dimensional complex manifold. We compute the Picard group of holomorphic line bundles on this loop space as an infinite dimensional complex Lie group with Lie algebra the first Dolbe…

2006-02-28abs ↗pdf ↗

We classify the simplest rational elements in a twisted loop group, and prove that dressing actions of them on proper indefinite affine spheres give the classical Tzitzéica transformation and its dual. We also give the group point of view of the Permutability Theorem, construct complex Tzitzéica transformations, and di…

2006-05-15abs ↗pdf ↗

Looped Transformers learn to implement multi-step gradient descent for in-context learning.

problem Understanding the learnability of multi-step algorithms in multi-layer Transformers.
method Training weight-sharing looped Transformers for in-context linear regression, proving gradient dominance condition for convergence.
result Looped Transformers implement multi-step preconditioned gradient descent, converging to global minimizer.

The 1-d Schrodinger flow on 2-sphere, the Gauss-Codazzi equation for flat Lagrangian submanifolds in C^n, and the space-time monopole equation are all examples of geometric soliton equations. The linear systems with a spectral parameter (Lax pair) associated to these equations satisfy the reality condition associated t…

2006-11-03abs ↗pdf ↗

Transformers learn to perform logistic regression in-context.

problem Understanding how transformers learn to perform specific tasks in-context.
method Constructed multi-layer transformers that perform in-context logistic regression through normalized gradient descent.
result Transformers can be trained to perform in-context logistic regression effectively.

We give a survey of the following six closely related topics: (i) a general method for constructing a soliton hierarchy from a splitting of a loop algebra into positive and negative subalgebras, together with a sequence of commuting positive elements, (ii) a method---based on (i)---for constructing soliton hierarchies …

2010-10-27abs ↗pdf ↗

We study Wilson-'t Hooft loop operators in a class of N=2 superconformal field theories recently introduced by Gaiotto. In the case that the gauge group is a product of SU(2) groups, we classify all possible loop operators in terms of their electric and magnetic charges subject to the Dirac quantization condition. We t…

2009-07-15abs ↗pdf ↗

The gradient flow of the Yang-Mills action acts pointwise on closed loops of gauge fields. We construct a topologically nontrivial loop of SU(2) gauge fields on S4 that is locally stable under the flow. The stable loop is written explicitly as a path between two gauge fields equivalent under a topologically nontrivial …

2010-08-06abs ↗pdf ↗

Develops Poisson structures on weak Sobolev loop spaces for integrable systems.

problem Analyzing integrable systems on low regularity loop spaces.
method Extending Mokhov's constructions to weak Sobolev spaces, constructing presymplectic and Poisson structures.
result Valid Poisson structures and deformations for weak Sobolev loops, extending Hamiltonian formalisms.

We associate a natural λλ-family (λR{0}λ\in \R \setminus \{0\} ) of flat Lagrangian immersions in $\C^n$ with non-degenerate normal bundle to any given one. We prove that the structure equations for such immersions admit the same Lax pair as the first order integrable system associated to the symmetric space $\frac{\U(n)…

2006-11-02abs ↗pdf ↗

We give, using an explicit expression obtained in [V. Jones, Ann. of Math. 126, 335 (1987)], a basic hypergeometric representation of the HOMFLY polynomial of (n,m)(n,m) torus knots, and present a number of equivalent expressions, all related by Heine's transformations. Using this result the (m,n)(n,m)(m,n)\leftrightarrow (n,m) s…

2014-01-31abs ↗pdf ↗

We construct a local action of the group of rational maps from S2S^2 to GL(n,C)GL(n,C) on local solutions of flows of the ZS-AKNS sl(n,C)sl(n,C)-hierarchy. We show that the actions of simple elements (linear fractional transformations) give local Bäcklund transformations, and we derive a permutability formula from different fact…

1998-05-18abs ↗pdf ↗

A human-in-the-loop ML framework for precision dosing reduces expert workload and removes bias.

problem High cost of data annotation and lack of appropriate data for ML models.
method Incorporates human experts into the model learning loop to improve interpretability and reduce bias.
result The approach learns interpretable rules from data and potentially lowers expert workload.

We show LLMs can be locally linear, enabling better control of activations.

problem Suboptimal control of LLM activations during generation.
method Model LLM inference as a linear dynamical system, compute feedback controllers using Jacobians, and adapt classical control theory.
result Robust, fine-grained control of LLM activations across models and tasks.

Enhanced decision-making through Dreamer's anticipatory trajectories and Online Decision Transformer.

problem Efficiently integrating world models with decision transformers.
method Combining Dreamer's trajectory forecasting with Online Decision Transformer's adaptive learning.
result Notable improvements in sample efficiency and reward maximization.

Multipath is among the major sources of errors in precise positioning using GPS and continues to be extensively studied. Two Fast Fourier Transform (FFT)-based detectors are presented in this paper as GPS multipath detection techniques. The detectors are formulated as binary hypothesis tests under the assumption that t…

2017-07-31abs ↗pdf ↗

Markov Chain Monte Carlo (MCMC) and Belief Propagation (BP) are the most popular algorithms for computational inference in Graphical Models (GM). In principle, MCMC is an exact probabilistic method which, however, often suffers from exponentially slow mixing. In contrast, BP is a deterministic method, which is typicall…

2016-05-29abs ↗pdf ↗

Proposes a new framework for learning image augmentations to improve classification performance.

problem Improving classification performance with a given class of predictors.
method Transformed Risk Minimization (TRM) framework that optimizes both predictive models and data transformations.
result Performance of TRM with SCALE algorithm compares favorably to prior methods on CIFAR10/100.

Study efficient pricing for barrier options in stochastic-volatility models with leverage correction.

problem Barrier options are sensitive to volatility dynamics, especially leverage, making accurate pricing difficult.
method Developed a class of continuous-path stochastic-clock volatility models and a systematic small-ρ expansion to incorporate leverage.
result Transform-only pricing formulas for barrier derivatives are fast and numerically stable, even for negative leverage.

Driven by the need for parallelizable hyperparameter optimization methods, this paper studies \emph{open loop} search methods: sequences that are predetermined and can be generated before a single configuration is evaluated. Examples include grid search, uniform random search, low discrepancy sequences, and other sampl…

2017-06-06abs ↗pdf ↗

The B^n(1)\hat B_n^{(1)}-hierarchy is constructed from the standard splitting of the affine Kac-Moody algebra B^n(1)\hat B_n^{(1)}, the Drinfeld-Sokolov B^n(1)\hat B_n^{(1)}-KdV hierarchy is obtained by pushing down the B^n(1)\hat B_n^{(1)}-flows along certain gauge orbit to a cross section of the gauge action. In this paper, we (1) u…

2019-12-15abs ↗pdf ↗

In this note we clarify the relation between extended world-sheet supersymmetry and generalized complex structure. The analysis is based on the phase space description of a wide class of sigma models. We point out the natural isomorphism between the group of orthogonal automorphisms of the Courant bracket and the group…

2005-02-15abs ↗pdf ↗

Our purpose is to explore, in the context of loop ensembles on finite graphs, the relations between combinatorial group theory, loops topology, loop measures, and signatures of discrete paths. We determine the distributions of the loop homotopy class, and of the first and second homologies, defined by the lower central…

2019-08-14abs ↗pdf ↗

We produce skew loops -- loops having no pair of parallel tangent lines -- homotopic to any loop in a flat torus or other quotient of R^n. The interesting case here is n=3. More subtly for any n, we characterize the homotopy classes that will contain a skew loop having a specified loop in the unit sphere as tangent ind…

2007-01-31abs ↗pdf ↗

Rational loops played a central role in Uhlenbeck's construction of harmonic maps into U(n) (chiral model in physics), and they are generated by simple elements with one pole and one zero constructed from Hermitian projections. It has been believed for long time that nilpotent loops should be added to generate rational…

2018-12-03abs ↗pdf ↗