Inserting label noise can improve model accuracy and fairness.
problem Improving model accuracy and fairness with noisy labels.
method Increasing label noise rates to balance and detect noisy instances.
result Inserting label noise can lead to more accurate and fair models.
A probabilistic framework generates an ensemble of models to defend against adversarial examples.
problem Defending against adversarial examples in deep learning models.
method Stochastic combinatorial ensembles with noise insertion decisions.
result It is possible to train models with limited adversarial attack transferability.
Transformer improves sequence generation with insertion and deletion phases.
problem Sequence generation challenges in machine translation.
method Insertion-Deletion Transformer with iterative insertion and deletion phases.
result Significant BLEU score improvement over insertion-only models.
Insertion Transformer generates sequences with flexible insertion orders.
problem Generating sequences with arbitrary orderings.
method Iterative, partially autoregressive model with insertion operations.
result Outperforms non-autoregressive approaches in translation tasks.
ID-ExpO fine-tunes neural networks for more faithful explanations.
problem Improving the faithfulness of explanations for complex machine learning models.
method Differentiable insertion/deletion metric-aware regularizers for optimization.
result Fine-tuned predictors produce more faithful explanations.
XL-Editor improves sentence post-editing using XLNet's variable-length insertion probability.
problem Post-editing sentences to refine generated text.
method XL-Editor trains XLNet to estimate variable-length insertion probabilities and apply post-editing operations.
result XL-Editor outperforms XLNet on text insertion and deletion tasks, and achieves significant style transfer improvements.
In this work we explore the use of metric index structures, which accelerate nearest neighbor queries, in the scenario where we need to interleave insertions and queries during deployment. This use-case is inspired by a real-life need in malware analysis triage, and is surprisingly understudied. Existing literature ten…
New framework for consistent submodular maximization with insertions and deletions.
problem Maintaining near-optimal solutions in a dynamic setting with insertions and deletions.
method Developed a general framework for fully dynamic submodular maximization, instantiated for cardinality and rank-k matroid constraints.
result First constant-factor approximations with sublinear consistency for both cardinality and rank-k matroid constraints.
New attacks improve privacy audits by analyzing model updates.
problem Improving privacy audits of AI models through sequence analysis.
method Developed SeMI attacks to identify target insertions in model sequences.
result SeMI attacks achieve higher power and tighter privacy audits.
Complex-valued neural networks improve wireless fingerprinting robustness.
problem Distinguish between devices sending the same message, robust against spoofing.
method Used complex-valued neural networks for supervised learning of fingerprints from wireless signals.
result Noise augmentation by adding white Gaussian noise leads to significant performance gains.
We propose a novel confidence scoring mechanism for deep neural networks based on a two-model paradigm involving a base model and a meta-model. The confidence score is learned by the meta-model observing the base model succeeding/failing at its task. As features to the meta-model, we investigate linear classifier probe…
The paper compares inserting and stretching points for grid refinement near critical points.
problem Decreased accuracy of finite difference methods near discontinuities.
method Comparison of inserting and stretching points for grid refinement near critical points.
result Proposes a new fast and simple stretching function.
Study shows resampling labels improves classifier performance in noisy data.
problem Balancing sample size vs label reliability in noisy data.
method Comparing different validation strategies and analyzing MNIST database with varying noise levels.
result Classifier performance declines with high incorrect labels, highlighting the importance of resampling.
CMT efficiently manages memory by inserting and querying memories in logarithmic time.
problem Managing large memory stores efficiently for quick access and updates.
method Designing a Contextual Memory Tree (CMT) that inserts and retrieves memories in logarithmic time.
result CMT improves classification algorithms and image-captioning tasks, demonstrating better computational efficiency.
Shared Keyboard design improves phase I clinical trials by borrowing information across doses.
problem Interim decisions based on current dose data may overlook signals from neighboring doses.
method Bayesian model-assisted design using Beta kernel process with kernel-weighted pseudo-counts.
result Significant improvements in identifying maximum tolerated dose and safety.
Dense neural networks learn efficiently with large datasets and noise.
problem Training neural networks with large, noisy datasets.
method Statistical mechanics and Monte Carlo simulations.
result Dense neural networks can handle large amounts of patterns and recognize patterns at high signal-to-noise ratios.
New formulas for feature importance tests in regression models.
problem Identifying important features in regression models.
method Established formulas for AUC criteria and proposed alternative metrics.
result Integrated Gradients (IG) performs nearly as well as Kernel SHAP (KS) but is faster.
IFH models graph generation with adjustable sequentiality.
problem Designing flexible graph generation models between one-shot and sequential approaches.
method Based on DDPM, IFH uses a node removal process to generate graphs with adjustable sequentiality.
result IFH models improve graph generation quality and efficiency compared to current methods.
Paper proposes P-ADMM for ADMM in distributed medical machine learning with differential privacy.
problem Privacy leakage in ADMM for distributed machine learning with sensitive data.
method Integrates Gaussian noise with linearly decaying variance to provide dynamic zCDP.
result P-ADMM achieves the same convergence rate as non-private ADMM while ensuring differential privacy.
agtboost speeds up gradient tree boosting with automatic complexity adjustment.
problem Speeding up and simplifying gradient tree boosting computations.
method Adaptive gradient tree boosting with automatic complexity adjustment and feature importance.
result Significant decrease in computation time and simplification of model complexity.
Study curvature and torsion from cross-ratios in discrete curves.
problem Define curvature and torsion for discrete curves using cross-ratios.
method Use Möbius invariant point-insertion-rule to construct circles and express torsion using cross-ratio.
result Discrete curvature and torsion defined using cross-ratios converge to smooth curvature and torsion as sampling density increases.
New algorithms update dynamic graph regression faster than existing methods.
problem Efficiently updating linear regression solutions for dynamic graphs.
method Subsampled randomized Hadamard transform and CountSketch.
result First sublinear update time randomized algorithms for dynamic graph regression.
We introduce backdrop, a flexible and simple-to-implement method, intuitively described as dropout acting only along the backpropagation pipeline. Backdrop is implemented via one or more masking layers which are inserted at specific points along the network. Each backdrop masking layer acts as the identity in the forwa…
In this paper we make two novel contributions to hierarchical clustering. First, we introduce an anomalous pattern initialisation method for hierarchical clustering algorithms, called A-Ward, capable of substantially reducing the time they take to converge. This method generates an initial partition with a sufficiently…
We give a proof of Ilmanen's lemma, which asserts that between a locally semi-convex and a locally semi-concave function it is possible to find a C1,1 function.
Inserts proximal mapping into deep networks for better regularization.
problem Effective regularization of deep learning models to handle adversarial perturbations and correlations between modalities.
method Proposes a new layer that directly produces regularized hidden layer outputs using proximal mapping.
result Outperforms state-of-the-art methods in robust temporal learning and multiview modeling.
Paper detects backdoors in neural networks using activation clustering.
problem Inserting backdoors into neural networks trained on untrusted data.
method Activation clustering to identify poisoned data.
result Demonstrates effectiveness in text and image classification.
We prove that the property of admitting no cosmetic crossing changes is preserved under the operation of forming certain satellites of winding number zero. We also define strongly cosmetic crossing changes and we discuss their behavior under the operation of inserting full twists in the strings of closed braids.
Paper introduces a technique to simplify RNN policies for better understanding and analysis.
problem Difficulty in explaining and analyzing RNN policies due to continuous-valued memory vectors and observation features.
method Quantized Bottleneck Insertion technique to learn finite representations of RNN vectors and features.
result Finite representations of RNN policies can be as small as 3 discrete memory states and 10 observations, improving interpretability.
The paper learns compact implicit surface maps from streaming data using an ensemble of sparse Gaussian processes.
problem Creating compact and accurate implicit surface maps from streaming range data.
method An ensemble of sparse Gaussian process experts, incrementally adjusted, trades-off between model complexity and prediction error.
result The approach learns compact and accurate implicit surface models comparable to or better than exact GP regression with subsampled data.
Graph neural networks perform well with pretraining even when new nodes and edges are added later.
problem Evaluation of graph neural networks' performance with dynamic graphs.
method Comparison of pretrained and retrained models in an experimental setup with dynamic graphs.
result Pretrained models maintain high accuracy on unseen nodes.
Bayesian framework inserts prior knowledge in RL for faster task solving.
problem Faster transfer learning across reinforcement learning tasks.
method Bayesian posterior distribution combining task-specific and prior knowledge.
result Significant speed ups achieved in maze solving.
The notion of a pseudoknot is defined as an equivalence class of knot diagrams that may be missing some crossing information. We provide here a topological invariant schema for pseudoknots and their relatives, 4-valent rigid vertex spatial graphs and singular knots, that is obtained by replacing unknown crossings or ve…
Study geodesic diameter on surfaces with special symmetry.
problem Geodesic diameter on surfaces with involutive isometry.
method Analyzes surfaces with specific symmetry properties.
result Determines geodesic diameter for these surfaces.
New approach shows backdoor attacks are indistinguishable from natural data features.
problem Defending against backdoor attacks in machine learning models.
method Developed a new primitive for detecting backdoor attacks based on the assumption that they correspond to the strongest feature in the training data.
result Backdoor attacks are indistinguishable from natural data features, making traditional detection methods ineffective.
Research characterizes critical points of scalar curvature functionals.
problem Characterizing critical points of scalar curvature functionals.
method Translation and analysis of a previous Russian paper.
result Provides insights into critical points of scalar curvature functionals.
Deep Partition Aggregation defends against poisoning attacks with provable certificates.
problem Adversarial poisoning attacks corrupt classifier test-time behavior.
method Deep Partition Aggregation (DPA) is an ensemble method using hash partitions and base models trained on these partitions.
result DPA can certify >= 50% of test images against over 500 poison image insertions on MNIST, and nine insertions on CIFAR-10.
Generative models can still learn from contaminated data, but with limitations.
problem How much contamination can generative models tolerate?
method Characterized robustness under contaminated enumerations, proving generation is achievable for all countable collections if contamination fraction converges to zero.
result Generation under contamination is achievable for all countable collections if contamination fraction converges to zero, but dense generation is strictly less robust.
Proof of Knot Entropy Conjecture for tube lattice polygons.
problem Proving exponential growth rate of knot polygons equals unknot polygons.
method Upper and lower bounds on polygon counts, braid insertions, and pattern theorems.
result Established the Knot Entropy Conjecture for tube lattice polygons.
KERMIT models sequences and pairs using a single neural network.
problem Efficient generative modeling for sequences and sequence pairs.
method Insertion-based approach using a single neural network.
result Unified model capable of matching or exceeding state-of-the-art performance.
A new method evolves point clouds using B-splines for smooth surfaces.
problem Evolution of smooth surfaces from discrete point clouds.
method Adaptive Lagrangian B-spline framework for geometric evolution.
result Efficient and accurate reproduction of surface evolution phenomena.
LGES speeds up causal discovery while maintaining accuracy.
problem Causal discovery from observational data with computational and accuracy limitations.
method LGES modifies GES by avoiding certain edge insertions, using prior knowledge, and leveraging interventional data.
result LGES outperforms GES in speed, accuracy, and robustness to misspecified knowledge.
If a contact form on a (2n+1)-dimensional closed contact manifold admits closed Reeb orbits, then its systolic ration is defined to be the quotient of (n+1)th power of the shortest period of Reeb orbits by the contact volume. We prove that every co-orientable contact structure on any closed contact manifold admits a co…
Improved language models with talking-heads attention.
problem Language model perplexity and quality issues.
method Added linear projections in multi-head attention.
result Better perplexities and quality in language tasks.
This paper is about interpolating minimal surfaces between two real analytic curves, a and b, each of which are simple real analytic curves, using the Björling-Schwarz formula in the domain where it is valid, changing the normal distributions on inital curves. We insert curves l1,...lL at specific locations and cla…
Paper defends LSTM-based text classification models from backdoor attacks.
problem Backdoor attacks in LSTM models cause misclassification of spam or malicious speech.
method Backdoor Keyword Identification (BKI) to identify and exclude poisoned samples.
result BKI method effectively mitigates backdoor attacks in various text classification datasets.
We consider hyperbolic 3-manifolds with either non-empty compact geodesic boundary, or some toric cusps, or both. For any such M we analyze what portion of the volume of M can be recovered by inserting in M boundary collars and cusp neighbourhoods with disjoint embedded interiors. Our main result is that this portion c…
Two price regimes identified in limit order books: close and far from quotes.
problem Understanding the distribution and behavior of limit orders in limit order books.
method Analysis of limit order book data in dimensions of price, time, lifetime, and volume.
result Identification of two distinct regimes in the limit order book: close and far from quotes.