arXiv subjects

Tuan A. Tran

Publications and source records attributed to Tuan A. Tran.

11 recordsLinked to original sources

arXiv2026

StructSAM: Structure- and Spectrum-Preserving Token Merging for Segment Anything Models

Recent token merging techniques for Vision Transformers (ViTs) provide substantial speedups by reducing the number of tokens processed by self-attention, often without retraining. However, their direct application to the Segment Anything Model (SAM) family is nontrivial: SAM's image encoder mixes windowed and global attention, and its mask decoder relies on dense, prompt-conditioned features for precise boundary prediction. We systematically evaluate representative token-merging methods on SAM and Medical SAM in a strict off-the-shelf setting, and find that existing destination-selection heuristics can erode boundaries and leak prompt information as merge rates increase. We propose \textbf{StructSAM}, a resolution-preserving merge-unmerge framework tailored to SAM. StructSAM computes a lightweight token-energy score from first-order feature gradients, uses grid-based flatness screening to protect boundary and prompt regions, and merges tokens within flat areas toward low-energy destinations with explicit token recovery. We further provide a spectral graph coarsening view showing that score-guided merging yields bounded Laplacian spectral distortion compared to random or window-restricted baselines. Across eight natural and medical benchmarks, StructSAM reduces encoder FLOPs by 25-30\% (up to 40\%+ with prompt-aware merging) with minor drops in mIoU/Dice, consistently outperforming ToMe, PiToMe, ToMeSD, VidToMe, and ALGM at the same compute.

cs.CV↗

arXiv2026

FOCA: Future-Oriented Conditioning for Data-Efficient Vision-Language-Action Adaptation

Vision-Language-Action (VLA) models enable general-purpose robotic control via large-scale multimodal pretraining, yet their effectiveness under few-shot imitation learning remains limited. We conduct a systematic stress test of state-of-the-art VLA models and show that performance degrades sharply as demonstrations are reduced, revealing a key weakness of existing adaptation strategies. To address this, we introduce FOCA, a future-oriented conditioning framework for data-efficient VLA adaptation. FOCA combines explicit prediction of task-grounded future interaction embeddings with implicit alignment to future goal observations, enabling long-horizon reasoning in latent space without pixel-level prediction. This formulation naturally supports action-free co-training with synthetic videos from video world models and can be interpreted as learning a future-conditioned value-like representation. Extensive experiments demonstrate FOCA achieves 95.7% success with 20 demonstrations on LIBERO, improves 7-12% on RoboCasa, and delivers up to 26% absolute gains on real robots, establishing a new state of the art in few-shot VLA adaptation.

cs.CV↗

arXiv2020

Accelerating Road Sign Ground Truth Construction with Knowledge Graph and Machine Learning

Having a comprehensive, high-quality dataset of road sign annotation is critical to the success of AI-based Road Sign Recognition (RSR) systems. In practice, annotators often face difficulties in learning road sign systems of different countries; hence, the tasks are often time-consuming and produce poor results. We propose a novel approach using knowledge graphs and a machine learning algorithm - variational prototyping-encoder (VPE) - to assist human annotators in classifying road signs effectively. Annotators can query the Road Sign Knowledge Graph using visual attributes and receive closest matching candidates suggested by the VPE model. The VPE model uses the candidates from the knowledge graph and a real sign image patch as inputs. We show that our knowledge graph approach can reduce sign search space by 98.9%. Furthermore, with VPE, our system can propose the correct single candidate for 75% of signs in the tested datasets, eliminating the human search effort entirely in those cases.

cs.AI↗

arXiv2003

The Fishnet as Anti-ferromagnetic Phase of Worldsheet Ising Spins

We identify the strong coupling fishnet diagram with a certain Ising spin configuration in the lightcone worldsheet description of planar TrΦ^3 field theory. Then, using a mean field formalism, we take the remaining planar diagrams into account in an average way. The fishnet spin configuration requires two mean fields ϕ,ϕ' where the fishnet diagram is the case ϕ=1,ϕ'=0. For general values of these fields, the system is then approximated as a light-cone quantized string with a field dependent effective string tension T_{eff}(ϕ,ϕ'). We also calculate the worldsheet energy density E(ϕ,ϕ'), and find the field values that minimize it in the presence of a transverse space infra-red cutoff ε>0. The criterion for string formation is that the tension in this minimum energy state remains non-zero as εgoes to 0. In the most simple-minded implementation of the mean field method, we find that the tension vanishes for weak and moderate coupling, but for very large coupling stays non-zero. A more elaborate treatment, taking temporal correlations into account, removes this ``phase transition'' and the string tension of the minimum energy state vanishes for all values of the coupling when εgoes to 0. Our mean field analysis thus suggests that the ``fishnet phase'' of TrΦ^3 theory is unstable, and there is no string formation for any value of the coupling. This is probably a reasonable outcome given the instability of the underlying theory.

hep-th↗

arXiv2002

BT Worldsheet for Supersymmetric Gauge Theories

We extend the Bardakci-Thorn (BT) worldsheet formalism to supersymmetric non-abelian gauge theory. Our method covers the cases of N =1,2,4 extended supersymmetry. This task requires the introduction of spinor valued Grassmann variables on the worldsheet analogous to those of the supersymmetric formulation of superstring theory. As in the pure Yang-Mills case, the worldsheet formalism automatically generates the correct quartic vertices from the cubic vertices. We also discuss coupling renormalization to one loop order.

hep-th↗

arXiv2001

Gauged Supergravities from Spherical Reductions

This dissertation is devoted to deriving the bosonic sectors of certain gauged supergravities in various dimensions from reducing eleven-dimensional supergravity, type IIA and type IIB supergravities in ten dimensions on certain spherical spaces. Explicit non-linear Kaluza-Klein ansätze for reductions of eleven-dimensional supergravity and of type IIA and type IIB supergravities on $S^n$ and $S^n\times T^m$ are presented. Knowing explicit non-linear ansätze is proven to be very useful in finding super Yang-Mills operators of gauge theories via AdS/CFT correspondence. We present a sample calculation which allows us to find a super Yang-Mills operator using a non-linear ansatz. Knowing non-linear ansätze is also useful for finding supergravity duals to certain twisted supersymmetric gauge theories. These supergravity solutions are branes wrapped on certain supersymmetric cycles. Some solutions, which are dual to gauge theories in three and five dimensions, are presented.

hep-th↗

arXiv2001

Supergravity duals of gauge theories from F(4) gauged supergravity in six dimensions

We construct supergravity solutions dual to twisted field theories that are the worldvolume theories of D4-branes wrapped on 2, 3-cycles, and NS-fivebranes on 3-cycles. We first obtain the solutions for the Romans' six-dimensional gauged supergravity theories and then up-lift them to ten dimensions. In particular, we find solutions for field configurations with either non-Abelian fields or B-fields being excited. One of these solutions, in the massless case, is up-lifted to the massless type IIA string theory. This is the first example of such a kind. The cases studied provide new examples of the AdS/CFT duality involving twisted field theories.

hep-th↗

arXiv2001

Supergravity duals of gauge field theories from SU(2) x U(1) gauged supergravity in five dimensions

We study the SO(4)-symmetric solution of the five-dimensional SU(2) x U(1) gauged N=4 supergravity theory obtained in [hep-th/0101202]. This solution contains purely magnetic non-Abelian and electric Abelian fields. It can be interpreted as a reduction of seven-dimensional gauged supergravity on a torus, which comes from type IIB supergravity on S^3. We also show how to obtain that solution from six-dimensional Romans' theory on a circle. We then up-lift the solution to massless type IIA supergravity. The dual gauge field theory is twisted and is defined on the worldvolume of a NS-fivebrane wrapped on S^3. Two other spatial directions of the NS-fivebrane are on a torus. In the IR limit it corresponds to a three-dimensional gauge field theory with two supercharges.

hep-th↗

arXiv1995

Electromagnetic detection of axions

Photon-to-axion conversions in the static electromagnetic fields are reconsidered in detail by using the Feynman diagram techniques. The differential cross sections are presented for the conversions in the presence of the electric field of the flat condenser as well as in the magnetic field of the solenoid. Based on our results a laboratory experiment for the production and the detection of the axions is described. This experiment will exploit the axion decay constant as well as the axion mass.

hep-ph↗

arXiv1994

Electromgnetic-gravitational cross-sections in external elctromagnetic fields

The classical processes: the conversion of photons into gravitons in the static electromagnetic fields are considered by using Feynman perturbation techniques. The differential cross sections are presented for the conversion in the electric field of the flat condesor and the magnetic field of the selenoid. A numerical evaluation shows that the cross sections may have the observable value in the present technical scenario.

astro-ph↗

arXiv1994

$\mbox{SU}(3)_L \otimes \mbox{U}(1)_N$ and $\mbox{SU}(4)_L \otimes \mbox{U}(1)_N$ gauge models with right-handed neutrinos

Pisano and Pleitez have introduced an interesting $\mbox{SU}(3)_C \otimes \mbox{SU}(3)_L \otimes \mbox{U}(1)_N$ gauge model which has the property that gauge anomaly cancellation requires the number of generations to be a multiple of 3. We consider generalizing that model to incorporate right-handed neutrinos. We find that there exists a non-trivial generalization of the Pisano-Pleitez model with right-handed neutrinos which is actually simpler than the original model in that symmetry breaking can be achieved with just three $\mbox{SU}(3)_L$ triplets (rather than 3 $\mbox{SU}(3)_L$ triplets and a sextet). We also consider a gauge model based on $\mbox{SU}(3)_C\otimes \mbox{SU}(4)_L \otimes \mbox{U}(1)_N$ symmetry. Both of these new models also have the feature that the anomalies cancel only when the number of generations is divisible by 3.

hep-ph↗