Training on mixed distributions improves test performance even when components are unrelated.
problem Improving test performance with mismatched training and test distributions.
method Analyzing mixture distributions with different training and test proportions.
result Distribution shift can be beneficial, improving test performance even when components are unrelated.
Paper optimizes training data distribution for better model performance across various deployment conditions.
problem Improving model accuracy when deployed with parameters far from training data.
method Developed adaptive algorithms based on bilevel or alternating optimization in the space of probability measures.
result Optimized training distributions lead to models with improved sample complexity and robustness to distribution shift.
Training machine learning (ML) models on large datasets requires considerable computing power. To speed up training, it is typical to distribute training across several machines, often with specialized hardware like GPUs or TPUs. Managing a distributed training job is complex and requires dealing with resource contenti…
Distributed Lion optimizes large model training by reducing communication costs.
problem Training large AI models efficiently with reduced communication costs.
method Adapted Lion optimizer for distributed training, using binary or lower-precision vectors for communication.
result Distributed Lion achieves comparable performance to standard optimizers but with significantly reduced communication bandwidth.
In-N-Out improves model robustness to out-of-distribution data.
problem Learning robust models with few in-distribution labeled examples.
method Pre-training with auxiliary information and self-training with pseudolabels.
result In-N-Out outperforms auxiliary inputs or outputs alone on both in-distribution and OOD error.
GANs learn distributions by matching low-degree moments.
problem Understanding when GANs learn the target distribution efficiently.
method Theoretical analysis and empirical observation of GAN training process.
result GANs can learn notable distributions by matching polynomially many low-degree moments.
Deep learning has led to tremendous advancements in the field of Artificial Intelligence. One caveat however is the substantial amount of compute needed to train these deep learning models. Training a benchmark dataset like ImageNet on a single machine with a modern GPU can take upto a week, distributing training on mu…
ShadowSync separates background synchronization for scalable distributed training.
problem Reducing synchronization overhead in distributed training for high scalability.
method Separates synchronization from training and runs it in the background.
result Achieves both high throughput and excellent model quality at scale.
New techniques improve distributed training with compressed gradients.
problem Gradient mismatch problem in local error feedback.
method Step-ahead error feedback and error averaging techniques.
result Our methods handle gradient mismatch and train faster than full-precision training.
This work proposes efficient classical training protocols for IQP circuits to train quantum generative models.
problem Training quantum generative models on industrially relevant probability distributions is challenging due to high computational cost.
method Developed protocols for classical training of IQP circuits, which are hard to sample but have efficient gradient computation.
result Classically trained IQP circuits can efficiently sample from target probability distributions, demonstrating practical quantum advantage.
We use GG distributions to optimize LLMs, reducing size and training time.
problem Lack of understanding in the statistical structure of LLMs.
method GG-based initialization, ACT, GCT.
result Smaller, faster models with minimal communication overhead.
The problem of detecting whether a test sample is from in-distribution (i.e., training distribution by a classifier) or out-of-distribution sufficiently different from it arises in many real-world machine learning applications. However, the state-of-art deep neural networks are known to be highly overconfident in their…
A distributed algorithm for training graph convolutional networks.
problem Training graph convolutional networks with sparse network topology and distributed agents.
method Formulate inference and optimization in a distributed scenario, propose a gradient descent procedure, and design communication topology.
result Convergence to stationary solutions of the GCN training problem under mild conditions.
The paper addresses the reliability of conformal prediction under covariate shift.
problem Ensuring reliable prediction sets under covariate shift.
method Derives upper bounds on training-conditional coverage.
result Offers PAC guarantees for conformal prediction methods.
Method diagnoses model performance under distribution shifts.
problem Understanding and improving model performance under distribution shifts.
method DIstribution Shift DEcomposition (DISDE) method to attribute performance drop to distribution shifts.
result Shows how model performance can be improved across different distribution shifts.
Ensemble approaches for uncertainty estimation have recently been applied to the tasks of misclassification detection, out-of-distribution input detection and adversarial attack detection. Prior Networks have been proposed as an approach to efficiently \emph{emulate} an ensemble of models for classification by paramete…
New method predicts model output distributions to improve data attribution.
problem Traditional training data attribution ignores randomness in model training.
method Distributional Training Data Attribution (d-TDA) using influence functions.
result Influence functions are effective and emerge naturally from d-TDA.
DistGP models multi-robot mapping with distributed Gaussian process learning.
problem Collaborative mapping by multiple robots with limited local data.
method Sparse Gaussian process with factorisation and distributed training via GBP.
result DistGP achieves superior accuracy and robustness compared to DiNNO.
Study on continuous sequence classification with distribution uncertainty.
problem Classifying continuous sequences with varying distribution uncertainty.
method Proposes distribution-free tests for three test designs: fixed-length, sequential, and two-phase tests.
result Error probabilities decay exponentially fast for all test designs.
Quantum models learn unitary actions on entangled states from product states.
problem Generalization to out-of-distribution data in quantum machine learning.
method Proved out-of-distribution generalization for learning unitary actions.
result Learned unitary actions on entangled states from product states.
Discriminatively trained neural classifiers can be trusted, only when the input data comes from the training distribution (in-distribution). Therefore, detecting out-of-distribution (OOD) samples is very important to avoid classification errors. In the context of OOD detection for image classification, one of the recen…
Public pretraining improves private model training even in extreme distribution shift scenarios.
problem Improving private model training accuracy in settings with large distribution shift.
method Empirical evaluation and theoretical explanation of public representations improving private training accuracy.
result Public representations can improve private training accuracy by up to 67% over private training from scratch in settings with large distribution shift.
Distributed machine learning (ML) can bring more computational resources to bear than single-machine learning, thus enabling reductions in training time. Distributed learning partitions models and data over many machines, allowing model and dataset sizes beyond the available compute power and memory of a single machine…
Neural networks are vulnerable to small adversarial perturbations. Existing literature largely focused on understanding and mitigating the vulnerability of learned models. In this paper, we demonstrate an intriguing phenomenon about the most popular robust training method in the literature, adversarial training: Advers…
Language models are generally trained on data spanning a wide range of topics (e.g., news, reviews, fiction), but they might be applied to an a priori unknown target distribution (e.g., restaurant reviews). In this paper, we first show that training on text outside the test distribution can degrade test performance whe…
Large-scale distributed training requires significant communication bandwidth for gradient exchange that limits the scalability of multi-node training, and requires expensive high-bandwidth network infrastructure. The situation gets even worse with distributed training on mobile devices (federated learning), which suff…
In this work, we propose a method to reject out-of-distribution samples which can be adapted to any network architecture and requires no additional training data. Publicly available chest x-ray data (38,353 images) is used to train a standard ResNet-50 model to detect emphysema. Feature activations of intermediate laye…
Causal influence measures for machine learnt classifiers shed light on the reasons behind classification, and aid in identifying influential input features and revealing their biases. However, such analyses involve evaluating the classifier using datapoints that may be atypical of its training distribution. Standard me…
A decision-theoretic bootstrapping method for robust uncertainty quantification.
problem Uncertainty in finite data sets and distributional shift between training and testing data.
method Partition data, train models, sample UQ subsets, define adversarial game, identify optimal mixed strategies.
result Optimal model mixtures and UQ estimates for robust uncertainty quantification.
New perspective on distribution shift helps make learning easier.
problem Learning with different training and target distributions.
method Formalizing and exploring Positive Distribution Shift (PDS).
result Distribution shift can be positive, making learning easier.
Combining self-training and contrastive learning improves performance under distribution shift.
problem Improving performance under distribution shift using unlabeled data.
method Combining self-training and contrastive learning techniques.
result Combined method achieves 3-8% higher accuracy than either approach independently.
SAPAG attacks distributed learning by reconstructing true training data from gradients.
problem Privacy attacks on distributed learning systems through gradients.
method SAPAG uses a Gaussian kernel-based gradient difference distance measure.
result SAPAG can reconstruct training data on various DNNs and at different training phases.
Domain randomization (DR) is a successful technique for learning robust policies for robot systems, when the dynamics of the target robot system are unknown. The success of policies trained with domain randomization however, is highly dependent on the correct selection of the randomization distribution. The majority of…
New methods reduce communication in distributed training for variational inequalities.
problem Reducing communication in distributed training for high-dimensional models.
method Distributed methods with compressed communication for solving variational inequalities.
result Theoretical guarantees and practical algorithms for compressed communication.
Distributed training of deep nets is an important technique to address some of the present day computing challenges like memory consumption and computational demands. Classical distributed approaches, synchronous or asynchronous, are based on the parameter server architecture, i.e., worker nodes compute gradients which…
RATIO improves neural network robustness and explainability.
problem Neural networks' lack of robustness to adversarial changes and uncertainty on out-distribution samples.
method RATIO: Adversarial Training on In- and Out-distribution.
result RATIO leads to robust models with reliable confidence estimates on out-distribution samples.
The paper optimizes training samples for image denoising across different noise levels.
problem Training a denoiser for all noise levels with uniform sample distribution.
method Derives a dual ascent algorithm for optimal sampling distribution.
result The algorithm converges to an optimal sampling distribution for deep neural networks.
FIRE method improves model performance in federated learning by penalizing fragmentation-induced covariate shifts.
problem Performance degradation in federated learning due to data fragmentation and covariate shift.
method FIRE method accumulates fragmentation-induced covariate shift divergences via approximate Fisher information and uses it as a per-fragment loss penalty.
result FIRE outperforms importance weighting and federated learning benchmarks by up to 5.3% on shifted validation sets.
TAET tackles long-tailed distributions in adversarial robustness.
problem Long-tailed distributions complicate adversarial robustness in real-world applications.
method TAET integrates an initial stabilization phase followed by a stratified equalization adversarial training phase.
result TAET achieves significant improvements in robustness and efficiency.
Adversarial training (AT) is among the most effective techniques to improve model robustness by augmenting training data with adversarial examples. However, most existing AT methods adopt a specific attack to craft adversarial examples, leading to the unreliable robustness against other unseen attacks. Besides, a singl…
Deep learning models frequently make incorrect predictions with high confidence when presented with test examples that are not well represented in their training dataset. We propose a novel and straightforward approach to estimate prediction uncertainty in a pre-trained neural network model. Our method estimates the tr…
New method trains neural samplers to sample from multi-modal distributions efficiently.
problem Mode-seeking behavior of reverse KL divergence hinders effective sampling from multi-modal target distributions.
method Minimizing reverse diffusive KL divergence along diffusion trajectories of model and target densities.
result Demonstrated enhanced sampling performance across various multi-modal distributions.
LSAM optimizes deep learning training with improved efficiency.
problem Inefficiency in distributed large-batch training with Sharpness-Aware Minimization (SAM).
method Integrates SAM's adversarial steps with an asynchronous distributed sampling strategy.
result Higher final accuracy compared to data-parallel SAM.
Deep learning with medical data often requires larger samples sizes than are available at single providers. While data sharing among institutions is desirable to train more accurate and sophisticated models, it can lead to severe privacy concerns due the sensitive nature of the data. This problem has motivated a number…
Study optimal ridge regularization for out-of-distribution prediction.
problem Optimal ridge regularization for predicting out-of-distribution data.
method Established conditions for optimal regularization under covariate and regression shifts, proving monotonic risk in data aspect ratio.
result Negative regularization can be optimal under shifts, even with isotropic or underparameterized training features.
ADB framework improves OOD generalization by increasing ID bias during training.
problem Machine learning models degrade on new data distributions.
method ADB framework introduces controlled statistical diversity during training.
result Higher in-distribution bias leads to better out-of-distribution generalization.
Reweighting training data to better represent new tasks.
problem Deploying machine learning models to new tasks is challenging due to training data distribution.
method Formulate an exponential tilt distribution shift model and learn train data importance weights to minimize KL divergence.
result The learned train data weights improve target performance evaluation, fine-tuning, and model selection.
Mobile edge learning is an emerging technique that enables distributed edge devices to collaborate in training shared machine learning models by exploiting their local data samples and communication and computation resources. To deal with the straggler dilemma issue faced in this technique, this paper proposes a new de…