Searcharxiv⌕ Search

arXiv subjects

Ken-Tye Yong

Publications and source records attributed to Ken-Tye Yong.

15 recordsLinked to original sources

How Many Training Labels Are Wrong? Auditing a Fitted Model through Gradient Allocation

How many labels in a training set are wrong, and which ones? The usual answers read a held-out set or the model's outputs. A held-out reading depends on how the held-out set was drawn, and an output-based reading fails once the model has fitted its wrong labels, because the loss of a fitted wrong label is small and the prediction agrees with it. We read the answer from the per-example gradients of a single trained model. Each example, together with its label, asks the network to change in a particular way. We normalise the magnitudes of its gradient into a gradient allocation and compare it with the mean allocation of every class. An example is flagged when relabelling it as its nearest other class makes its allocation fit better than under its own label, and the share of flagged examples is the reading of the noise rate. The rule has no threshold and needs no clean data, no held-out set and no further training. After the model has fitted 99-100% of its wrong labels, the allocation still separates wrong from right labels with an AUC of 0.94-0.96 on the noisiest human labels of CIFAR-10N, where a feature-space margin falls to 0.79-0.84 and the loss, the gradient norm and the self-influence of an example to 0.73-0.76. On such a fitted model, readings of the noise rate from predictions or feature margins fall to 0-6% at true rates of 8-58% on CIFAR-10 with ViT-S, while the audit reads 54-96% of the true rate, and at least 93% of the examples it counts carry wrong labels. On a briefly trained model, before the wrong labels are fitted, the reading lies within 1.7 percentage points of the true noise rate on CIFAR-10 from 0% to 58% noise. Confident learning, which trains four further models, separates wrong labels about as well (AUC 0.925-0.998 against 0.900-0.995) and mostly reads the noise rate more closely, at a lower precision. An audit of 5,000 examples takes 14 to 27 seconds on one GPU.

cs.LG↗

Routing in Gradient Space: Balanced Usage Is Not Expert Specialization

Sparse expert models can distribute traffic evenly while still grouping incompatible training signals within the same experts. We study routing as a gradient-partitioning problem and introduce gradient-aligned routing (GAR), whose load-normalized router objective rewards grouping observations with aligned gradients. On five multi-task text-classification mixtures, we compare GAR with task-loss-only routing, gradient-combination and gradient-conflict methods, and load-balancing losses. With a fully trainable RoBERTa backbone and classification-head experts, GAR has the highest aggregate validation accuracy, 1.07 percentage points above task-loss-only routing. With frozen DeBERTa and Qwen3-1.7B backbones and low-rank adapter experts, it again ranks first, 1.10 points above task-loss-only routing, with better-balanced expert load and higher gradient-mass purity, the share of each expert's gradient-norm mass from its dominant task; the load-balancing losses flatten load further but leave this purity near its task-loss-only level. Top-1 routing, trainable full-parameter feed-forward network (FFN) experts, and a larger backbone also show positive aggregate gains. In component ablations, the objective's cross-observation alignment terms alone retain GAR's accuracy, and GAR keeps its accuracy when observations are formed without task identity; supervising the router with task labels raises task-expert association without adding accuracy. The results distinguish expert-load balance from gradient-based routing organization and show the predictive value of gradient-informed routing in multi-task text classification.

cs.LG↗

Why Does Train-Validation Separation Emerge? Gradient-Allocation Dynamics in Pretrained Fine-Tuning

During fine-tuning, training accuracy often keeps rising after validation accuracy has levelled off. We explain this growing train-validation gap with a support-transfer account: early updates use structure shared by many examples, whereas later fitting relies increasingly on narrow structure whose benefit transfers less to held-out data, so per-example gradient allocations differentiate as the gap grows. We measure this with two readouts computed from fixed training probes alone, the pooled dispersion $D_{\mathrm{den}}$ of normalized gradient magnitudes and its within-class part $D_{\mathrm{within}}^{\mathrm{wt}}$, and derive sufficient conditions under which allocation dispersion, cross-entropy separation, and the accuracy gap grow together. In a constructed feature hierarchy, raising the sample-private feature share from .3 to .7 raises the final accuracy gap from .185 to .527, and homogenizing per-example gradient magnitudes reduces the gap mainly after substantial fitting (median full-dose changes of -4.91 to -11.79 percentage points when it starts at the three later training stages). In 90 fine-tuning runs of three language backbones on six tasks, both readouts are positively associated with the accuracy gap in every run, and $D_{\mathrm{den}}$ remains positive in 89 of 90 runs after linear detrending; 20 vision runs show the same association.

cs.LG↗

When Reviews Gate Access: Collusion, Sabotage and the Scale of Peer Review in the AI Submission Surge

AI tools have helped push submissions to the largest machine-learning venues to tens of thousands per cycle. Venues have answered with rules that tie access to review: per-author caps, mandatory reciprocal reviewing, reviewer eligibility earned through accepted papers, and assignment based on reviewer bids. Under these rules a reviewer is also a competitor of the authors. We study two deviations that the rules make attractive. Sabotage lowers a competitor's score. Collusion raises an ally's score. We build an agent-based conference calibrated to public statistics of NeurIPS and ICLR and run 6,220 simulated conferences. Four results stand out. First, the two deviations do different harm and amplify each other. Collusion changes who is accepted, sabotage erodes what scores say about quality, and rings that may also sabotage gain more than rings that only collude. Second, under random assignment the chance that collusion changes a decision is set by one exposure index, $E = Rm/A$, the expected number of allies among a paper's reviewers. Growth therefore protects random assignment, but bidding removes the protection. In a pool of 3,000 reviewers a 20-member ring turns 0.8% of its papers into acceptances under random assignment and 71% when 30% of allies' bids are granted. Third, when agents choose their allies every cycle, caps at the inner layers of the social brain keep collusion nearly harmless. Without a cap agents ally with half of their area and accepted value falls by 35%. Fourth, reviewer eligibility tied to acceptances turns sabotage into a weapon of incumbents, and a longer eligibility window widens the gap between incumbents and newcomers only when sabotage is possible.

cs.MA↗

Generative machine learning for protein design

Designing proteins with specific biological functions is of considerable interest for both the development of new therapeutics and the engineering of enzymes, yet achieving the intended activity remains challenging. Generative artificial intelligence (AI) has emerged as a powerful tool to explore the vast space of possible designs by learning from protein sequence and structural data. Here, we review 59 generative approaches across six protein design tasks, including sequence generation, backbone generation, structure-conditioned sequence design, joint sequence-structure design, motif scaffolding, and binder and antibody design. We outline common generative approaches and protein representations and discuss how structural constraints and functional information are utilized during generation. This is followed by a discussion of the computational and experimental approaches used to evaluate designed proteins and what such assessments reveal about their properties. Finally, we discuss current limitations in designing proteins beyond well-represented families and incorporating experimental feedback into the design process. We expect that rapid advances in generative modeling will enable more reliable design of a wider range of protein structures and functions over the coming years.

cs.LG↗

Axial von Mises Framework for Tensor-Dependent Second-Harmonic Generation

We derive closed-form relations between projected fibril-axis statistics and tensor-dependent second-harmonic generation (SHG) harmonics using a single axial von Mises population. Under coherent emission within domains and detected intensity addition across domains, each angular harmonic factors into a local tensor coefficient and an axial circular moment. Parallel-analyzer detection contains second, fourth and sixth harmonics; ideal equal-weight analyzer-free detection contains second and fourth harmonics. The resulting formulas identify geometry-specific harmonic zeros, sensitivity loss and ambiguous inversion of a secondary fixed-angle ratio. In particular, the analyzer-free fourth-harmonic zero at $ρ=1/3$ does not imply loss of all orientation sensitivity. Prepared-cornea forward/reverse recordings illustrate extraction of empirical analyzer-free harmonics. In the bundle ROI, second harmonics are resolved in both sweeps, while their normalized amplitudes differ and fourth harmonics remain unresolved. An exploratory central ROI from the same recordings yields a different forward/reverse profile. Within the same operator-identified tissue, low-signal space coverage and apparent fibre alignment vary slightly across recordings. These observations support angular-measurement feasibility while leaving quantitative normalized-shape agreement unestablished. The thermal extension illustrates consequences of assumed kinetics rather than calibrated tissue trajectories. The contribution is an analytical interpretation of SHG harmonics with explicit conditions for physical inference; the measurements do not establish absolute susceptibility ratio $ρ$, axial concentration $κ$, or experimental separation of thermal disorder from degradation.

physics.optics↗

Backpropagated Output Momentum: Relocating Optimizer History from Parameters to Task Space

Optimizer momentum is usually stored as a parameter-sized moving average of past gradients, which makes history costly and fixes each past signal in the coordinates in which it was computed. We introduce Backpropagated Output Momentum (BOM), which instead stores a compact moving average of prediction errors at the model output and reprojects that history through the current network at every step. A batch-level analysis characterizes the information retained and omitted by this relocation, while the implementation preserves the current supervised gradient and can replace the first-moment component of several adaptive optimizers. As a plug-in for momentum-based optimizers, including ones that already compress their state, BOM reduces parameter-shaped optimizer state by 49.7-99.8% in three compositions and, averaged over three language backbones, paired step time by 4.0%. It also improves mean validation performance across language and vision fine-tuning, by 1.42 points in the primary five-task comparison. Language and vision pretraining studies, together with matched mechanism controls, further test the construction across output spaces and model scales.

cs.LG↗

Q-BridgeNet: A Quantization Network for Cross-Lingual Sign Language Translation

Most sign language translation (SLT) methods focus on isolated native sign-spoken pairs (e.g., American Sign Language - English). Extending language-specific SLT models to multilingual translation would improve accessibility by enabling communication across diverse sign and spoken language communities. However, existing multilingual SLT approaches still struggle to learn a unified model that minimizes cross-lingual conflicts while capturing shared cross-lingual semantics and preserving language-specific variations across different sign languages. Therefore, we propose Q-BridgeNet, a unified framework for multilingual SLT that jointly mitigates cross-lingual conflicts across both the sign language and spoken language sides. On the sign language side, Q-BridgeNet learns discrete Q-units via adaptive segmentation and residual vector quantization: a shared base codebook provides language-agnostic semantic primitives, while language-specific residual codebooks refine heterogeneous signing semantics. On the spoken language side, a multilingual LLM is fine-tuned to operate in the Q-unit space, leveraging cross-lingual priors to enable a unified SLT model. Experiments on PHOENIX14T, How2Sign, and CSL-Daily show that Q-BridgeNet effectively mitigates cross-lingual conflicts, achieving state-of-the-art performance on native sign-spoken pairs while also demonstrating strong generalization to non-native pairs. Our source code is publicly available at: https://github.com/FengLiQ/Q-BridgeNet

cs.CL↗

A Unified Spectrum for Turbulence in Microfluidic Flow

We present a predictive master spectrum describing turbulence-like flows in microfluidic systems. Extending Pao's viscous-range closure, the model introduces (i) an adaptive inertial-range slope dependent on measurable dimensionless numbers and (ii) a physics-specific cutoff that captures entropy-producing sinks such as electrokinetic forcing, compliant walls, active stresses, and interfacial tension. This formulation unifies turbulence regimes -- electrokinetic, active, interfacial, and compressible -- within one compact expression. Comparison with reported data reproduces both spectral slopes and dissipation cutoffs while requiring only global observables (velocity, viscosity, Taylor microscale, and forcing strength). The framework provides a design-level predictive tool for turbulent microflows prior to computationally heavy DNS or CFD.

physics.flu-dyn↗

Universal Quartic Scaling Law for Kerr-Type Interactions: Projection-Law Factorization Across Nonlinear Quantum Platforms

We present a rigorous derivation and numerical validation of a universal projection-law factorization for quartic nonlinear coupling rates across physically distinct platforms. The central result is that observable Kerr-type interactions -- self-Kerr, cross-Kerr, and cross-phase modulation -- factorize into a dimensionless projection coefficient and an intrinsic quartic energy scale. This structure follows from canonical quantization of a quartic interaction projected onto a finite normal-mode basis. We state this result as a formal proposition with clearly enumerated assumptions, provide a proof sketch based on second quantization of the anharmonic potential, and enumerate the domain of validity. We then implement the scaling law as a lightweight computational toolkit (UEFT-Designer) with platform-specific kernels for superconducting circuits, photonic microcavities, and epsilon-near-zero (ENZ) structures. A complete, step-by-step worked example for a superconducting quarton device -- with full uncertainty propagation -- predicts a cross-Kerr rate $χ/2π= 361\pm 13\,\mathrm{MHz}$, agreeing with the independently measured value of $366\pm 0.5\,\mathrm{MHz}$ \cite{Ye2025} to within 1.4\%. A formal error-propagation analysis establishes that the dominant uncertainty is the Josephson-energy extraction uncertainty ($\sim\!2\%$), not the geometric projection factor. Cross-platform validation against five independent experiments spanning eight orders of magnitude in coupling strength confirms the universality of the factorization to within reported experimental uncertainties. The ghost-sector spectral correction introduced in earlier versions is reframed as an optional phenomenological self-energy model with no mandatory role in the validated scaling law.

quant-ph↗

Computation and Communication Efficient Federated Unlearning via On-server Gradient Conflict Mitigation and Expression

Federated Unlearning (FUL) aims to remove specific participants' data contributions from a trained Federated Learning model, thereby ensuring data privacy and compliance with regulatory requirements. Despite its potential, progress in FUL has been limited due to several challenges, including the cross-client knowledge inaccessibility and high computational and communication costs. To overcome these challenges, we propose Federated On-server Unlearning (FOUL), a novel framework that comprises two key stages. The learning-to-unlearn stage serves as a preparatory learning phase, during which the model identifies and encodes the key features associated with the forget clients. This stage is communication-efficient and establishes the basis for the subsequent unlearning process. Subsequently, on-server knowledge aggregation phase aims to perform the unlearning process at the server without requiring access to client data, thereby preserving both efficiency and privacy. We introduce a new data setting for FUL, which enables a more transparent and rigorous evaluation of unlearning. To highlight the effectiveness of our approach, we propose a novel evaluation metric termed time-to-forget, which measures how quickly the model achieves optimal unlearning performance. Extensive experiments conducted on three datasets under various unlearning scenarios demonstrate that FOUL outperforms the Retraining in FUL. Moreover, FOUL achieves competitive or superior results with significantly reduced time-to-forget, while maintaining low communication and computation costs.

cs.LG↗

Molecular Dynamics Simulations Reveal PolyQ-Length-Dependent Conformational Changes in Huntingtin Exon-1: Implications for Environmental Co-Solvent Modulation of Aggregation-Prone States

Huntington's disease (HD) is caused by CAG-repeat expansion in HTT, which lengthens the polyglutamine (polyQ) tract in huntingtin (HTT) and promotes misfolding and aggregation. While polyQ-length-dependent aggregation is well established, the atomistic conformational dynamics preceding aggregation remain less defined. Here we perform all-atom molecular dynamics simulations of HTT exon-1 constructs containing the N17 domain, polyQ tracts of clinically relevant lengths (Q21, wildtype; Q40, adult onset threshold; Q70, juvenile onset), and the polyproline (polyP) region. Multi-copy simulations (four chains) were run for 100 ns in explicit SPC/E water using the OPLS-AA force field. We quantified radius of gyration (Rg), solvent-accessible surface area (SASA), root-mean-square deviation (RMSD), and intra-protein hydrogen bonds as proxies for conformational expansion and aggregation propensity. PolyQ expansion drove progressive increases in Rg and SASA, consistent with more extended, solvent-exposed ensembles. We further tested organic co-solvents (methanol, hexane, trichloroethylene; 0.5 to 1.0 M), which modulated these landscapes in a solvent-dependent manner. Trichloroethylene induced marked expansion in Q21 and Q40, whereas methanol produced mild compaction in Q21. To our knowledge, this is the first MD study to systematically examine co-solvent effects on HTT exon-1 conformational dynamics. Although limited sampling precludes definitive mechanistic conclusions, the observed trends suggest that hydrophobic co-solvents can bias HTT exon-1 toward more expanded ensembles, motivating computational studies of gene-environment modulation in HD.

cs.CE↗

Unified Effective Field Theory for Nonlinear and Quantum Optics

Predicting phenomena that mix few-photon quantum optics with strong field nonlinear optics is hindered by the use of separate theoretical formalisms for each regime. We close this gap with a unified effective field theory valid for frequencies lower than the material-dependent cutoff set by the band gap, plasma frequency, or similar scale. The action couples the electromagnetic gauge field to vector polarisation modes. An isotropic potential generates the optical susceptibilities, while a higher-dimension axion-like term captures magnetoelectric effects; quantisation on the Schwinger-Keldysh contour with doubled BRST ghosts preserves gauge symmetry in dissipative media. One-loop renormalisation-group equations reproduce the measured dispersion of the third-order susceptibility from terahertz to near-visible frequencies after matching a single datum per material. Real-time dynamics solved with a matrix-product-operator engine yield good agreement with published results for GaAs polariton cavities, epsilon-near-zero indium-tin-oxide films and superconducting quarton circuits. The current formulation is limited to these 1-D geometries and sub-cut-off frequencies; higher-dimensional or above-cut-off phenomena will require additional degrees of freedom or numerical methods.

physics.optics↗

Advancements in Programmable Lipid Nanoparticles: Exploring the Four-Domain Model for Targeted Drug Delivery

Programmable lipid nanoparticles, or LNPs, represent a breakthrough in the realm of targeted drug delivery, offering precise spatiotemporal control essential for the treatment of complex diseases such as cancer and genetic disorders. In order to provide a more modular perspective and a more balanced analysis of the mechanism, this review presents a novel Four-Domain Model that consists of Architecture, Interface, Payload, and Dispersal Domain. We explored the dynamical equilibrium between LNPs components and the surroundings throughout their destiny, from formulation to release. On the basis of this, we delve deep into manufacturing challenges, scalability issues, and regulatory hurdles, associated with the clinical translation of LNP technology. Within the framework focusing on the programmability in each domain, we prioritized patient-centric factors like dosing regimens, administration techniques, and potential consequences. Notably, this review expands to innovative anatomical routes, such as intranasal and intraocular administration, offering a thorough examination of the advantages and disadvantages of each route. We also offered a comprehensive comparison between artificial LNPs and natural exosomes in terms of functionality, biocompatibility, and therapeutic potential. Ultimately, this review highlights the potential of programmable LNPs to evolve into more intelligent, naturally integrated systems, achieving optimal biocompatibility and functionality.

q-bio.BM↗

Phase-controllable growth of ultrathin 2D magnetic FeTe crystals

Two-dimensional (2D) magnets with intrinsic ferromagnetic/antiferromagnetic (FM/AFM) ordering are highly desirable for future spintronics devices. However, the synthesis of 2D magnetic crystals, especially the direct growth on SiO2/Si substrate, is just in its infancy. Here, we report a chemical vapor deposition (CVD)-based rational growth approach for the synthesis of ultrathin FeTe crystals with controlled structural and magnetic phases. By precisely optimizing the growth temperature (Tgrowth), FeTe nanoplates with either layered tetragonal or non-layered hexagonal phase can be controlled with high-quality. The two controllable phases lead to square and triangular morphologies with a thickness down to 3.6 and 2.8 nm, respectively. More importantly, transport measurements reveal that tetragonal FeTe is antiferromagnetic with a Neel temperature (TN) about 71.8 K, while hexagonal FeTe is ferromagnetic with a Curie temperature (TC) around 220 K. Theoretical calculations indicate that the ferromagnetic order in hexagonal FeTe is originated from a concomitant lattice distortion and the spin-lattice coupling. This study represents a major step forward in the CVD growth of 2D magnetic materials on SiO2/Si substrates and highlights on their potential applications in the future spintronic devices.

cond-mat.mtrl-sci↗