A new method selects training samples for fine-tuning using validation set inference.
problem Selecting training examples for fine-tuning with limited target data.
method Invert train-validation roles; select samples affecting most predictions.
result Our method achieves lower test log-loss than state-of-the-art approaches.
Fine-tuning with pre-training data improves performance.
problem Limited training data for tasks.
method Theoretical analysis of excess risk bound and selection of pre-training data subset.
result Improvement in generalization performance with pre-training data.
Selective joint fine-tuning improves deep learning with limited data.
problem Insufficient labeled training data for deep learning tasks.
method Joint fine-tuning of shared convolutional layers between source and target tasks using selected training images.
result Improves classification accuracy by 2% - 10% on multiple visual classification tasks.
FisherSFT selects informative examples to fine-tune LLMs efficiently.
problem Adapting large language models to new domains efficiently.
method Selects examples maximizing information gain using Hessian of log-likelihood.
result Empirically demonstrates improved performance with reduced computational cost.
LOFT separates subspace rotation and transformation for orthogonal fine-tuning.
problem Conflating subspace rotation and transformation in orthogonal fine-tuning.
method LOFT explicitly separates subspace rotation and transformation, using task-aware support selection.
result LOFT recovers principal-subspace orthogonal adaptation and improves efficiency-performance trade-off.
Optimizes sparse fine-tuning for privacy in neural networks.
problem Performance gap between DP-SGD and non-private fine-tuning.
method Optimization-based approach using private gradient information for selecting trainable weights.
result Our selection method leads to better prediction accuracy compared to existing approaches.
In2Core selects a coreset for efficient LLM fine-tuning with reduced data.
problem Costly fine-tuning of large language models due to extensive parameters and data requirements.
method Analyzes model gradients to estimate training sample influence, optimizing for efficiency.
result Achieves similar performance with 50% of training data using In2Core.
SelMix fine-tunes pre-trained models to optimize non-decomposable objectives.
problem Optimizing non-decomposable performance measures for practical applications.
method Selective mixup fine-tuning of pre-trained models.
result SelMix significantly improves performance for various non-decomposable objectives.
SMART-FAN-Lasso fine-tunes neural networks for high-dimensional nonparametric regression.
problem Fine-tuning neural networks for high-dimensional nonparametric regression with variable selection.
method Source-model-augmented residual tuning (SMART) framework for neural Lasso.
result SMART-FAN-Lasso achieves statistical acceleration over single-task learning under precise conditions.
EVA adapts LoRA for faster, more efficient fine-tuning.
problem Fast and efficient fine-tuning of large models for specific tasks.
method EVA uses directions capturing most activation variance for initialization, maximizing gradient signal and reducing parameters.
result EVA achieves faster convergence and higher average scores across tasks, reducing parameters.
This paper re-evaluates hyperparameters for fine-tuning pre-trained models.
problem Current hyperparameter settings for fine-tuning are often ad-hoc and fixed.
method Empirical evaluation of learning rate, batch size, and momentum for fine-tuning.
result Optimal hyperparameters are not only dataset-dependent but also sensitive to domain similarity.
GeLoRA optimizes LoRA fine-tuning by dynamically adjusting ranks based on intrinsic dimensionality.
problem Efficient fine-tuning of large language models with limited computational resources.
method GeLoRA computes intrinsic dimensionality to adaptively select LoRA ranks, balancing expressivity and efficiency.
result GeLoRA consistently outperforms recent baselines within the same parameter budget on multiple tasks.
Fine-tuning neural networks to guarantee performance on specific examples can also introduce incorrect inputs.
problem Ensuring reliable performance of neural networks on specific examples.
method Using SMT solvers to fine-tune ReLU neural networks to guarantee outcomes on a finite set of particular examples.
result Fine-tuning can introduce incorrect inputs that trigger unexpected performance.
Reduces annotation costs in medical imaging by 50%.
problem Challenges in creating large annotated datasets for medical imaging.
method Integrates active learning and transfer learning into a single framework.
result Reduces annotation efforts by at least half.
Two methods for model adaptation compared; fine-tuning outperforms Best-of-N in realizable settings.
problem Comparing methods for adapting large language models to new tasks.
method Supervised fine-tuning vs. Best-of-N approach.
result Supervised fine-tuning outperforms Best-of-N in realizable settings.
The condition number predicts efficient information encoding in neural units, aiding model fine-tuning.
problem Efficient information encoding in neural units for various tasks and input modalities.
method Linking the condition number to the log-volume scaling factor and entropy of the output distribution.
result High condition number indicates efficient encoding, reducing overall information transfer.
A new principle for optimizer selection improves training speed and performance.
problem Finding the best optimizer hyperparameters for faster training.
method Formulate optimizer selection as maximizing the expected drop rate in loss, treating gradients and updates as signals and an optimizer as a causal filter.
result Greedy optimizer selection yields stable and effective momentum rules.
A trainable gate optimizes neural network selection and pruning.
problem Optimizing neural networks for specific tasks.
method Introduces a trainable gate function to make discrete selection problems differentiable.
result Efficiently optimizes arbitrary neural networks across various tasks.
Improves classifier performance in multi-stage selection processes.
problem Difficulty in training classifiers in multi-stage selection processes due to varying sample sizes and information.
method Multi-Stage Transfer Learning (MSGTL) approach that uses knowledge from simpler classifiers trained in early stages to improve later stages.
result MSGTL outperforms other transfer learning methods in real-world selection process data.
MallowsPO enhances LLM fine-tuning with a dispersion index of human preferences.
problem Lack of diversity in human preferences in DPO.
method Developed a dispersion index based on Mallows' theory to characterize preference diversity.
result Demonstrated improved performance in various tasks using the dispersion index.
LORENZA improves LLM fine-tuning efficiency and generalization.
problem Improving robustness and generalization of LLMs under hardware constraints.
method AdaZo-SAM and LORENZA, combining Adam and SAM with zeroth-order estimation and randomized SVD.
result LORENZA achieves better generalization and reduced memory consumption compared to existing methods.
Efficiently compress neural networks with MUSCO method.
problem Compression of deep neural networks.
method Iterative approach alternating low-rank factorization with rank selection and fine-tuning.
result Improves compression rate while maintaining accuracy.
Cost-effective method improves and re-purposes pre-trained GANs by fine-tuning class-embeddings.
problem Fine-tuning BigGANs from scratch is impractical due to instability and high computational cost.
method Fine-tuning only the class-embedding layer of pre-trained GANs.
result Significantly improved realism and diversity of samples, re-purposed for new tasks, and de-biased or improved diversity.
S2D selectively decays large singular values to improve quantization of neural activations.
problem Large activation outliers in transformer models cause accuracy drops during quantization.
method Selective Spectral Decay (S2D) that surgically regularizes only the largest singular values. result Significantly reduces activation outliers and produces well-conditioned representations.
Generative Adversarial Networks improve trading strategy performance.
problem Optimizing trading strategies in a competitive market.
method Conditional Generative Adversarial Networks (cGANs) for strategy calibration and combination.
result cGANs provide outperformance over traditional techniques in generating alpha.
MixFT re-partitions data into sub-domains for better TSFM fine-tuning.
problem Improving zero-shot forecasting for new time series domains.
method MixFT re-divides data using Bayesian mixtures into homogeneous sub-domains for separate fine-tuning.
result MixFT outperforms per-dataset fine-tuning methods.
Curriculum learning improves model training efficiency and performance.
problem Expensive training of reasoning models using human or synthetic data.
method Autocurriculum, where models use their own performance to decide training problems.
result Autocurriculum reduces training costs and improves model performance.
RAFT fine-tunes models using high-quality samples to align them with human preferences.
problem Aligning generative models with human ethics and preferences.
method RAFT selects high-quality samples, discards undesired behavior, and fine-tunes the model on filtered samples.
result RAFT improves model performance in reward learning and automated metrics.
Spectrum selectively trains LLMs based on SNR to save resources.
problem Efficiently training large language models with limited resources.
method Targeting layer modules based on SNR for selective training.
result Spectrum achieves similar performance to full fine-tuning but with reduced VRAM usage.
Enhances adversarial example transferability by fine-tuning existing examples.
problem Adversarial examples are often overfit to a source model, limiting black-box transferability.
method Intermediate Level Attack (ILA) fine-tunes adversarial examples on a pre-specified layer of the source model.
result ILAs achieve high transferability to target models without knowledge of their architecture.
Bayesian principles improve neural additive models for better feature selection and uncertainty.
problem Lack of calibrated uncertainties and feature selection in neural additive models.
method Augmenting NAMs with Bayesian principles to provide credible intervals, feature selection, and interaction ranking.
result Improved performance on tabular datasets and real-world medical tasks.
DPTA improves CIL by adapting PTMs with dual prototypes.
problem Catastrophic forgetting in incremental learning with pre-trained models.
method Dual-Prototype Network with Task-wise Adaptation (DPTA).
result DPTA consistently outperforms recent methods by 1\%-5\% on multiple benchmarks.
We formalize AURC and develop estimators for SC systems.
problem Evaluation of SC systems' performance.
method Formal statistical formulation, Monte Carlo methods, plug-in estimators.
result Plug-in estimators are consistent, with low bias and bounded MSE.
BWS selects best window subsets for efficient data pruning.
problem Challenges in selecting subsets of large datasets for neural network training.
method Best Window Selection (BWS) by choosing optimal window intervals from ordered sample scores.
result BWS outperforms other methods across various selection ratios and datasets.
New measure FTC quantifies how much a ReLU network can fine-tune.
problem Analyzing memorization capacity in fine-tuned neural networks.
method Defined Fine-Tuning Capacity (FTC) for additive fine-tuning of ReLU networks.
result Upper and lower bounds on FTC for 2 and 3-layer ReLU networks.
New method uses limited labeled data and multiple starts to adapt models across domains.
problem Accurate predictions in target domain with few labeled data.
method Fine-tuning from multiple adaptive starts, extending UDA methods.
result Minimax-optimal target performance with limited labeled target data.
New method quantifies uncertainty in fine-tuned LLMs using LoRA ensembles.
problem Uncertainty in fine-tuned LLMs and how to trust their predictions.
method Posterior approximations using low-rank adaptation ensembles.
result Unexpected retention of acquired knowledge during fine-tuning in overfitting regime.
BERT fine-tuning is unstable due to optimization issues, not forgetting or dataset size.
problem Stability of fine-tuning BERT-based models across different random seeds.
method Analysis of BERT, RoBERTa, and ALBERT fine-tuned on GLUE datasets, identifying optimization difficulties as the cause of instability.
result Fine-tuning instability is due to optimization difficulties leading to vanishing gradients, not forgetting or dataset size.
Selective privacy framework for recommender systems.
problem Adversarial nature of collaborative filtering to user privacy.
method Selective privacy preserving (SP2) framework with two-step process.
result Users can control what ratings are publicly shared, leading to more accurate recommendations.
Improved code translation by preserving structure with composed fine-tuning.
problem Improving code translation accuracy with unlabeled code outputs.
method Pre-trained denoiser to capture output structure, composed fine-tuning to fine-tune predictor.
result Composed fine-tuning significantly improves generalization over standard fine-tuning.
Study uses zero-shot models to forecast mortality rates globally.
problem Forecasting mortality rates without task-specific fine-tuning.
method Two state-of-the-art foundation models (TimesFM and CHRONOS) and traditional/machine learning methods were evaluated.
result CHRONOS outperformed traditional methods for shorter-term forecasts, but TimesFM consistently underperformed.
New Bregman chord divergences simplify distance selection in machine learning.
problem Selecting appropriate distances for machine learning tasks.
method Extend Bregman divergences with two scalar parameters.
result Simplified distance selection with asymptotic generalization of Bregman divergences.
The paper introduces a Hessian-based method to improve generalization in fine-tuned deep neural networks.
problem Improving generalization in fine-tuned deep neural networks, especially in noisy conditions.
method PAC-Bayesian analysis to identify a Hessian-based distance measure, proving generalization bounds, and developing an algorithm with a generalization error guarantee.
result Hessian-based distance measure correlates well with observed generalization gaps and can match the scale of these gaps in practice.
Fine-tuning LLMs improves capability but harms safety, study finds.
problem Balancing capability and safety in LLM fine-tuning.
method Theoretical framework and numerical experiments for two safety-aware fine-tuning strategies.
result Characterization of fundamental limits of safety-capability trade-off in LLM fine-tuning.
Fine-tuning harms in-context learning, but restricting updates to the value matrix improves zero-shot performance.
problem Fine-tuning harms in-context learning, reducing zero-shot performance on unseen tasks.
method Theoretical analysis of linear attention models, identifying conditions for degraded few-shot performance.
result Restricting updates to the value matrix improves zero-shot performance while preserving in-context learning.
SpotTune adapts fine-tuning strategies per instance for improved transfer learning.
problem Improving transfer learning performance with deep neural networks.
method Adaptive fine-tuning approach using policy networks to decide whether to use pre-trained or fine-tuned layers.
result SpotTune outperforms traditional fine-tuning on 12 out of 14 standard datasets and achieves highest scores on Visual Decathlon.
Measures consistency of tabular LLM predictions under fine-tuning multiplicity.
problem Conflicting predictions from fine-tuned tabular LLMs.
method Local stability measure in embedding space.
result Probabilistic guarantees on prediction consistency under multiplicity.
Self-play fine-tuning improves diffusion models for text-to-image generation.
problem Plateauing performance of diffusion models after data saturation.
method Self-play fine-tuning (SPIN-Diffusion) using competition among model versions.
result Significantly improved model performance and human preference alignment.