ICML 2026 Papers

Low-Rank Adaptation (LoRA) has gained popularity as a fine-tuning approach for Large Language Models (LLMs) due to its low resource requirements and good performance. While numerous studies have investigated improving LoRA serving efficiency by serving multiple LoRAs concurrently, existing methods assume that a wide range of LoRA adapters are available for serving. In our work, we conduct extensive empirical studies to show that current LoRA training paradigms do not efficiently utilize hardware resources and incur high overhead to obtain a performant LoRA adapter. Leveraging these insights, we propose PLoRA, which automatically orchestrates concurrent LoRA fine-tuning jobs under given hardware and model constraints and develops performant kernels to improve training efficiency. Across a range of LLMs and LoRA configurations, PLoRA improves training throughput by up to 12.8x and reduces the overall fine-tuning makespan by up to 7.52x compared to existing approaches.

Foundation models have achieved remarkable success, yet their growing parameter counts pose significant computational and memory challenges. Low-rank factorization offers a promising route to reduce training and inference costs, but the community lacks a stable recipe for training models from scratch using exclusively low-rank weights while matching performance of the dense model. We demonstrate that Large Language Models (LLMs) can be trained from scratch using exclusively low-rank factorized weights for all non-embedding matrices without auxiliary "full-rank" guidance required by prior methods. While native low-rank training often suffers from instability and loss spikes, we identify uncontrolled growth in the spectral norm (largest singular value) of the weight matrix update as the dominant factor. To address this, we introduce **Spectron: Spectr**al renormalization with orthogonalizati**on**, which dynamically bounds the resultant weight updates based on the current spectral norms of the factors. Our method enables stable, end-to-end factorized training with negligible overhead. Finally, we establish compute-optimal scaling laws for natively low-rank transformers, demonstrating predictable power-law behavior and improved inference efficiency relative to dense models. Our code is available at [https://github.com/Pauljanson002/spectron](https://github.com/Pauljanson002/spectron)

Post-training quantization at the 2-bit level enables low-cost deployment and inference acceleration for large language models (LLMs). Scalar quantization (SQ) and vector quantization (VQ) are two primary quantization methods, however, the former suffers from significant performance degradation, and the latter incurs computational and storage overhead. We propose UniSVQ, a unified 2-bit quantization framework that bridges scalar and vector quantization by parameterizing codewords as an affine transform of integer lattices. This structure preserves compatibility with optimized integer kernels while retaining much of VQ's flexibility. We further introduce a data-driven block-wise fine-tuning strategy to directly minimize quantization reconstruction error. Extensive experiments across multiple LLM families and zero-shot benchmarks demonstrate that UniSVQ consistently outperforms state-of-the-art SQ methods and achieves performance comparable to advanced VQ methods, while providing higher inference throughput.

Driven by the imperative to leverage privacy-sensitive data scattered across decentralized devices, federated fine-tuning has emerged as a vital paradigm for adapting large language models without compromising data privacy. Yet, its practical efficacy is bottlenecked by severe client resource heterogeneity. Existing truncation-based methods typically couple the transmitted rank with the trainable rank, which (i) under-utilizes bandwidth on communication-rich but compute-limited clients and (ii) exacerbates truncation-induced gradient drift. To address this, we propose FedHera, a resource-decoupled framework that explicitly differentiates information reception from gradient optimization. FedHera employs a spectrum-preserving allocation strategy to maximize the transfer of global knowledge (via high-rank singular values) within bandwidth limits, irrespective of training constraints. Furthermore, we introduce a prefix-gating mechanism that utilizes the downloaded high-capacity basis as a frozen reference to guide local updates, thereby minimizing the optimization gap caused by aggressive truncation. Extensive experiments under different heterogeneous settings show that FedHera improves stability and accuracy over state-of-the-art baselines.

Geometric properties of Transformer weights, particularly the unembedding matrix, have been widely useful in language model interpretability research. Yet, their utility for estimating downstream performance remains unclear. In this work, we systematically investigate the relationship between model performance and the unembedding matrix geometry, particularly its effective rank. Our experiments, involving a suite of 108 OLMo-style language models trained under controlled variation, reveal several key findings. While the best-performing models often exhibit a high effective rank, this trend is not universal across tasks and training setups. Contrary to prior work, we find that low effective rank does not cause late-stage performance degradation in small models, but instead co-occurs with it; we find adversarial cases where low-rank models do not exhibit saturation. Moreover, we show that effective rank is strongly influenced by pre-training hyperparameters, such as batch size and weight decay, which in-turn affect the model's performance. Lastly, extending our analysis to other geometric metrics and final-layer representation, we find that these metrics are largely aligned, but none can reliably predict downstream performance. Overall, our findings suggest that the model's geometry, as captured by existing metrics, primarily reflects training choices rather than performance.

Low-Rank Adaptation (LoRA) is the most widely adopted method for fine-tuning large language models. Notably, LoRA is inherently overparameterized: multiple pairs of low-rank factors can yield the same adapted weight matrix. We show—both theoretically and empirically—that these pairs exhibit significantly different condition numbers. As a result, converging to different loss minimizers directly impacts the convergence rate of LoRA. Building on this observation, we introduce Balanced Low-Rank Adaptation (BaLoRA), a variant of LoRA that projects iterates onto a balanced manifold. This manifold improves the conditioning of the loss landscape while preserving the adapted matrix. The projection step is computationally lightweight and integrates seamlessly into existing fine-tuning pipelines. Empirically, BaLoRA converges faster than standard LoRA and achieves superior performance across a range of fine-tuning tasks.

Meta-learning methods perform well on new within-distribution tasks but often fail when adapting to out-of-distribution target tasks, where transfer from source tasks can induce negative transfer. We propose a causally-aware Bayesian meta-learning method, by conditioning task-specific priors on precomputed latent causal task embeddings, enabling transfer based on mechanistic similarity rather than spurious correlations. Our approach explicitly considers realistic deployment settings where access to target-task data is limited, and adaptation relies on noisy (expert-provided) pairwise judgments of causal similarity between source and target tasks. We provide a theoretical analysis showing that conditioning on causal embeddings controls prior mismatch and mitigates negative transfer under task shift. Empirically, we demonstrate reductions in negative transfer and improved out-of-distribution adaptation in controlled simulations and a real-world clinical prediction setting for cross-disease transfer, where causal embeddings align with underlying clinical mechanisms; we include the judgments from a medical expert in the clinical prediction task and obtain improved performance in predictions of unseen diseases.

Video Diffusion Transformers (VDiTs) represent the state of the art in video generation but remain constrained by the quadratic complexity of dense self-attention. To address this attention bottleneck, we analyze the pre-softmax matrix ($QK^\top$) and reveal two key properties: (1) video attention exhibits an effective low-dimensional structure with rapid singular-value decay, and (2) real motion induces hybrid spatio-temporal patterns rather than rigid ``spatial vs. temporal'' layouts. Guided by these observations, we propose Video-SVD, a training-free and plug-and-play acceleration method that does not modify the original network parameters. Video-SVD learns checkpoint-adaptive orthogonal bases offline and, at inference time, replaces expensive dense attention computation with lightweight online subspace projection and basis composition. To preserve high fidelity, Video-SVD further employs layer-shared dual-stream residual modules to recover fine-grained content details and positional information. Across HunyuanVideo and Wan2.1 backbones, Video-SVD achieves significant end-to-end speedups while maintaining high visual quality, reaching 1.92$\times$ on HunyuanVideo, 1.75$\times$ on Wan2.1-1.3B, and 1.79$\times$ on Wan2.1-14B.

Implicit Neural Representation (INR) has emerged as a powerful paradigm for continuous MRI reconstruction. However, standard self-supervised INR requires time-consuming optimization from scratch for each scan, hindering clinical deployment. This work presents IPOD, a Reference-Free Meta-Learning framework designed to learn generalized parameter initializations for INR directly from undersampled data. Distinct from conventional meta-learning that relies on fully-sampled ground truth, IPOD operates in an inverse-problem-driven manner, leveraging diverse reconstruction tasks with varying sampling patterns to capture a robust prior. Furthermore, we introduce an adaptive meta-update strategy modulated by task-specific performance to ensure optimal parameter distribution for diverse anatomical structures. Extensive experiments demonstrate that IPOD provides a superior initialization that enables rapid adaptation and achieves high-fidelity reconstruction across various imaging protocols, significantly outperforming existing INR baselines. By eliminating the dependence on reference images, IPOD offers a scalable and efficient solution for a wide range of imaging inverse problems. Code and data available at: https://github.com/zhn00310/RFML4MRI

Vision-language models (VLMs) have progressed rapidly with large-scale high-quality data and adaptation strategies, yet remain brittle under real-world corruptions, where both visual recognition and language-grounded reasoning degrade. Beyond cascaded image restoration, a natural alternative is parameter-efficient adaptation that aligns corrupted features with clean references; however, Euclidean alignment alone is not semantics-preserving and can even harm downstream reasoning. We attribute this to a semantic misalignment gap, where features become geometrically closer while drifting off the in-distribution support on which multimodal reasoning is calibrated. To address this, we propose Manifold-Adversarial Adapters (MAA), parameter-efficient layer-wise modules for a frozen vision encoder that explicitly steer corrupted features back onto the clean in-distribution manifold rather than merely shrinking feature-space distance. MAA combines paired feature self-distillation with a token-level adversarial manifold constraint to prevent off-manifold semantic shortcuts. At inference, only the adapters are retained, enabling single-stage robustness with negligible overhead and avoiding the latency and semantic drift of restoration pipelines. Across benchmarks and corruption settings, MAA consistently improves performance over strong baselines.

Test-time adaptation (TTA) offers a compelling remedy for machine learning (ML) models that degrade under domain shifts, improving generalisation *on-the-fly* with only unlabelled samples. This flexibility suits real deployments, yet conventional evaluations unrealistically assume unbounded processing time, overlooking the accuracy-latency trade-off. As ML increasingly underpins latency-sensitive and user-facing use-cases, temporal pressure constrains the viability of adaptable inference; predictions arriving too late to act on are futile. We introduce *Tempora*, a framework for evaluating TTA under this pressure. It consists of temporal scenarios that model deployment constraints, evaluation protocols that operationalise measurement, and time-contingent utility metrics that quantify the accuracy-latency trade-off. We instantiate the framework with three such metrics: (1) *discrete* utility for asynchronous streams with hard deadlines, (2) *continuous* utility for interactive settings where value decays with latency, and (3) *amortised* utility for budget-constrained deployments. By applying Tempora to 11 TTA methods, we find that *rank instability* persists across 750+ temporal evaluations spanning diverse datasets, models, and hardware platforms; *i.e.,* conventional rankings do not predict rankings under temporal pressure. The highest-utility method varies with the shift and temporal pressure, with no clear winner. By enabling systematic evaluation across diverse temporal constraints for the first time, Tempora reveals when and why rankings change, offering practitioners a lens for method selection and researchers a target for deployable adaptation. Code: https://github.com/sudotensor/tempora.

Gradual domain adaptation (GDA) aims to mitigate domain shift by progressively adapting models from the source domain to the target domain via intermediate domains. However, real intermediate domains are often unavailable or ineffective, necessitating the synthesis of intermediate samples. Flow-based models have recently been used for this purpose by interpolating between source and target distributions. Notably, their training typically relies on sample-based log-likelihood estimation, which can discard useful information and thus degrade GDA performance. The key to addressing this limitation is constructing the intermediate domains via samples directly. To this end, we propose an $\underline{\text{E}}$ntropy-regularized $\underline{\text{S}}$emi-dual $\underline{\text{U}}$nbalanced $\underline{\text{O}}$ptimal $\underline{\text{T}}$ransport (E-SUOT) framework to construct intermediate domains. Specifically, we reformulate flow-based GDA as a Lagrangian dual problem and derive an equivalent semi-dual objective that circumvents the need for likelihood estimation. However, the dual problem leads to an unstable min–max training procedure. To alleviate this issue, we further introduce the entropy regularization to convert it into a more stable sequential optimization procedure. Based on this, we propose a novel GDA training framework and provide theoretical analysis in terms of stability and generalization. Finally, extensive experiments are conducted to demonstrate the efficacy of the E-SUOT framework.

Long input sequences are central to in-context learning, document understanding, and multi-step reasoning in Large Language Models (LLMs). However, the quadratic attention cost of Transformers makes inference memory-intensive and slow. While context distillation (CD) can transfer information into model parameters, per-prompt distillation is impractical due to training costs and latency. To address these limitations, we propose Doc-to-LoRA (D2L), a lightweight hypernetwork that meta-learns to perform approximate CD within a single forward pass. Given an unseen prompt, D2L generates a LoRA adapter for a target LLM, enabling subsequent queries to be answered without re-consuming the original context, reducing latency and KV-cache memory consumption during target LLM inference. On a long-context needle-in-a-haystack task, D2L successfully learns to map contexts into adapters that store the needle information, achieving near-perfect zero-shot accuracy at sequence lengths exceeding the target LLM’s native context window by more than 4x. On real-world QA datasets with limited compute, D2L outperforms standard CD while significantly reducing peak memory consumption and update latency. We envision that D2L can facilitate rapid adaptation of LLMs, opening up the possibility of frequent knowledge updates and personalized chat behavior. Code and checkpoints are available at https://github.com/SakanaAI/doc-to-lora.

In the data selection problem, the objective is to choose a small, representative subset of data that can be used to efficiently train a machine learning model. Sener and Savarese [ICLR 2018] showed that, given an embedding representation of the data and suitable geometric assumptions, heuristics based on $k$-center clustering can be used to perform data selection. This perspective was further explored by Axiotis et. al. [ICML 2024], who proposed a data selection approach based on $k$-means clustering and sensitivity sampling. However, these methods rely on the assumption that the dataset exhibits intrinsic geometric structure that can be effectively captured by clustering, whereas many modern datasets instead possess global algebraic structure that is better exploited by low-rank approximation or principal component analysis. In this paper, we introduce a new data selection framework based on low-rank approximation and residual-based sampling, formulated through the lens of row subset selection and loss-preserving coreset construction. Given an embedding representation of the data satisfying mild regularity conditions, which can be interpreted as algebraic or angular notions of Lipschitz continuity, we show that it is possible to select a weighted subset of $\tilde{O}\left(k + \frac{1}{\varepsilon^2}\right)$ data points whose average loss approximates the average loss over the full dataset within a $(1+\varepsilon)$ relative error, up to an additive $\varepsilon \Phi_k$ term, where $\Phi_k$ denotes the optimal rank-$k$ approximation cost of the embedding matrix. We complement these theoretical guarantees with empirical evaluations, demonstrating that on a range of real-world datasets, our data selection approach achieves improved performance over prior strategies based on uniform sampling or clustering-based sensitivity sampling.

A central challenge in large language model (LLM) editing is capability preservation: methods that successfully change targeted behavior can quietly game the editing proxy and corrupt general capabilities, producing degenerate behaviors reminiscent of proxy/reward hacking. We present CrispEdit, a scalable and principled second-order editing algorithm that treats capability preservation as an explicit constraint, unifying and generalizing several existing editing approaches. CrispEdit formulates editing as constrained optimization and enforces the constraint by projecting edit updates onto the low-curvature subspace of the capability-loss landscape. At the crux of CrispEdit is expressing capability constraint via Bregman divergence, whose quadratic form yields the Gauss–Newton Hessian exactly and even when the base model is not trained to convergence. We make this second-order procedure efficient at the LLM scale using Kronecker-factored approximate curvature (K-FAC) and a novel matrix-free projector that exploits Kronecker structure to avoid constructing massive projection matrices. Across standard model-editing benchmarks and safety unlearning tasks, CrispEdit achieves high edit success while keeping capability degradation below 1% on average across datasets, significantly improving over prior editors.

We study neural networks with trainable low-degree rational activation functions and show that they are more expressive and parameter-efficient than modern piecewise-linear and smooth activations such as ELU, LeakyReLU, LogSigmoid, PReLU, ReLU, SELU, CELU, Sigmoid, SiLU, Mish, Softplus, Tanh, Softmin, Softmax, and LogSoftmax. For an error target of $\varepsilon>0$, we establish approximation-theoretic separations: Any network built from standard fixed activations can be uniformly approximated on compact domains by a rational-activation network with only $\mathrm{poly}(\log\log(1/\varepsilon))$ overhead in size, while the converse provably requires $\Omega(\log(1/\varepsilon))$ parameters in the worst case. This exponential gap persists at the level of full networks and extends to gated activations and transformer-style nonlinearities. In practice, rational activations integrate seamlessly into standard architectures and training pipelines, allowing rationals to match or outperform fixed activations under identical architectures and optimizers.

Optimal transport (OT) finds a least-cost transport plan between two probability distributions. Unlike standard OT, which infers unstructured pointwise mappings, low-rank optimal transport explicitly constrains the rank of the transport plan to infer latent structure. This improves statistical robustness, yields sharper parametric rates for estimating Wasserstein distances, and generalizes $K$-means to co-clustering. However, these advantages come at the cost of a non-convex and NP-hard optimization problem. We introduce Transport Clustering, an algorithm to compute a low-rank OT plan that reduces low-rank OT to a clustering problem on correspondences obtained from a full-rank transport registration step. We prove that this reduction yields polynomial-time, constant-factor approximation algorithms for low-rank OT: specifically, a $(1+\gamma)$ approximation for negative-type metrics, a $(1+\gamma+\sqrt{2\gamma}\,)$ approximation for kernel costs, and a $(1+\gamma+\rho)$ approximation for general metrics satisfying the triangle inequality. Here, $\gamma \in [0,1]$ is the cost ratio of the optimal full-rank to low-rank solutions, and $\rho\in[0,1]$ is an asymmetry coefficient on the cluster variances. Numerically, Transport Clustering outperforms existing solvers on synthetic benchmarks and large-scale datasets.

Federated Learning (FL) with Low-Rank Adaptation (LoRA) has become a standard for privacy-preserving LLM fine-tuning. However, existing personalized methods predominantly operated under a restrictive Flat-Model Assumption: they addressed client-side *statistical heterogeneity* but treated the model as a monolithic block, ignoring the *functional heterogeneity* across LLM layers. We argue that these two statistical (horizontal) and functional (vertical) dimensions, are *orthogonal in source yet coupled in interaction*, implying that the optimal depth of parameter sharing is functionally dependent on client similarity. To address this, we propose **FedTreeLoRA**, a framework employing tree-structured aggregation for fine-grained, layer-wise alignment. By dynamically constructing an aggregation hierarchy, FedTreeLoRA allows clients to share broad consensus on shallow ’trunks‘ while progressively specializing on deep ‘branches'. Experiments on NLU and NLG benchmarks demonstrate that FedTreeLoRA significantly outperforms state-of-the-art methods by effectively reconciling generalization and personalization.

Long-context language modeling is commonly framed as a scalability challenge of token-level attention, yet local-to-global information structuring remains largely implicit in existing approaches. Drawing on cognitive theories of discourse comprehension, we propose HiCI (Hierarchical Construction--Integration), a hierarchical attention module that constructs segment-level representations, integrates them into a shared global context, and broadcasts both to condition segment-level attention. We validate HiCI through parameter-efficient adaptation of LLaMA-2 (7B and 13B) and Qwen3-8B with only $\sim$4--5\% additional parameters, extending context to 100K/64K tokens for LLaMA-2-7B/13B and to 48K tokens for Qwen3-8B. Across language modeling, retrieval, and instruction-following benchmarks, HiCI yields consistent improvements over strong baselines, including matching proprietary models on topic retrieval and surpassing GPT-3.5-Turbo-16K on code comprehension. These results demonstrate the effectiveness of explicit hierarchical structuring as an inductive bias for long-context modeling.