SI learning is effective for unbalanced data, showing resilience to object dependencies.
problem Learning from unbalanced data in multi-instance learning.
method Analysis of SI learning objective with rich classifier families, focusing on unbalanced data.
result Unbalanced data improves resilience of SI method to object dependencies, especially in neural networks.
Proposes a new MIL formulation using infinitely many shapelets.
problem Weakness of single shapelet classifiers and lack of theoretical guarantee for multiple shapelets.
method Formulates a new MIL approach with infinitely many shapelets and provides an efficient algorithm.
result Empirical study shows effectiveness in MIL and Shapelet Learning.
This paper introduces collective counterfactual explanations for groups of instances in classification models.
problem Understanding how classification models make decisions for groups of instances.
method Novel Mathematical Optimization models to find collective counterfactual explanations that minimize total perturbation cost.
result Detects critical features for entire dataset classification and handles outliers.
Proposes a Siamese NN for algorithm selection focusing on alike performing instances.
problem Lack of effective meta-features for algorithm selection via meta-learning.
method Siamese Neural Network architecture with 'Algorithm-Performance Personas' concept.
result Proposed metric outperforms standard performance metrics in training sample selection.
A new method for instance segmentation using a single network.
problem Precisely detecting and delineating objects in images.
method Proposes a solution based on DCME technique, using a single segmentation network.
result Reduces computational cost by repurposing the network encoder for classification.
We present an instance segmentation algorithm trained and applied to a CCTV recording of beef cattle during a winter finishing period. A fully convolutional network was transformed into an instance segmentation network that learns to label each instance of an animal separately. We introduce a conceptually simple framew…
Paper uses financial news for stock trend forecasting using deep multiple instance learning.
problem Forecasting stock trends from financial news articles.
method Developed a flexible and adaptive multi-instance learning model for bags of instances (financial news articles) on trading days.
result Outstanding trend prediction accuracy compared to state-of-the-art approaches.
Multiple instance learning (MIL) is a variation of supervised learning where a single class label is assigned to a bag of instances. In this paper, we state the MIL problem as learning the Bernoulli distribution of the bag label where the bag label probability is fully parameterized by neural networks. Furthermore, we …
Labeling data for classification requires significant human effort. To reduce labeling cost, instead of labeling every instance, a group of instances (bag) is labeled by a single bag label. Computer algorithms are then used to infer the label for each instance in a bag, a process referred to as instance annotation. Thi…
End-to-end method learns geometry and appearance for multi-view object detection.
problem Challenges in multi-view object detection, including viewpoint, lighting, and scale variability.
method Jointly learns multi-view geometry and warping for robust cross-view object detection.
result Superior performance compared to baselines on a new street-level panorama data set.
Improves key instance detection in MIL models by using neural network inversion with sparseness constraint.
problem Limited key instance detection performance in attention-based deep MIL models due to skewed attention scores.
method Sparse network inversion with a sparseness constraint incorporated into neural network inversion, solved by proximal gradient method.
result Significantly improved key instance detection performance while maintaining bag-level prediction performance.
A method to approximate instance-dependent label noise using instance-confidence embedding.
problem Real-world label noise that depends on individual instances.
method Variational approximation with instance embedding to capture instance-specific label corruption.
result ICE method effectively approximates instance-dependent noise and detects ambiguous instances.
Crowdsourcing has become very popular among the machine learning community as a way to obtain labels that allow a ground truth to be estimated for a given dataset. In most of the approaches that use crowdsourced labels, annotators are asked to provide, for each presented instance, a single class label. Such a request c…
Paper proposes a method to generate instance labels from weakly supervised data.
problem Weakly supervised instance labeling in medical image analysis.
method Uses multiple instance learning (MIL) and knowledge distillation to generate instance-level predictions.
result Significantly outperforms state-of-the-art MIL methods in instance-level prediction.
Panoptic-DeepLab achieves state-of-the-art results in panoptic segmentation.
problem Panoptic segmentation challenges in computer vision.
method Bottom-up, single-shot approach with dual-ASPP and dual-decoder structures.
result Panoptic-DeepLab sets new state-of-the-art results on Cityscapes benchmarks.
M4L-JMF tackles multi-typed objects learning, improving on M3L.
problem Learning from multi-typed objects with diverse features and labels.
method Joint matrix factorization to encode and factorize multi-typed bags and their instances.
result M4L-JMF outperforms existing methods on benchmark datasets.
InstaNAS searches for a distribution of architectures to improve performance and reduce latency.
problem Finding a single architecture that represents the whole dataset with high diversity and variety.
method InstaNAS uses a controller trained to search for a 'distribution of architectures' that assigns each input sample a domain expert architecture.
result InstaNAS achieves up to 48.8% latency reduction without compromising accuracy.
SymNet uses neural models to solve RMDPs efficiently.
problem Limited effectiveness of generalized policies in RMDPs.
method SymNet trains shared parameters for an RDDL domain, converts instances to graphs, uses relational neural models for node embeddings, and scores actions based on embeddings.
result SymNet policies outperform random and state-of-the-art deep reactive policies on nine RDDL domains.
RISE framework unifies and improves time series learning with missing data.
problem Learning from time series with missing data.
method RISE framework unifies and improves time series learning with missing data.
result RISE instances always benefit from encoders that learn representations for numerical values.
New method improves learning from multiple correlated data trajectories.
problem Learning from multiple correlated data trajectories without mixing assumptions.
method Hellinger localization framework for maximum likelihood estimation.
result Instance-optimal bounds that scale with full data budget under broad conditions.
A deep RL approach generates counterfactual instances efficiently.
problem Efficient generation of counterfactual instances for large datasets and diverse models.
method Deep reinforcement learning to optimize counterfactual instances in a single forward pass.
result Model-agnostic and scalable counterfactual generation.
A new SSL method using whitening of latent-space features.
problem Efficiency and effectiveness of self-supervised representation learning.
method Proposes a new loss function based on whitening of latent-space features, avoiding the need for negatives and asymmetric networks.
result Improves efficiency and effectiveness of SSL by avoiding the need for negatives and asymmetric networks.
SurvBeNIM explains survival models using neural networks.
problem Explaining predictions of survival models.
method Extends Beran estimator with neural importance functions.
result Improved explanation of survival models compared to existing methods.
Novel framework for efficient entity alignment labeling.
problem Efficient labeling of entity alignments in knowledge graphs.
method Active and passive learning strategies to select informative instances.
result Passive learning achieves comparable performance to active learning.
Proposes a new method for MIR using kernel mean embeddings.
problem Multiple instance regression (MIR) where bags contain multiple instances with a single label.
method Computes kernel mean embeddings of predicted label distributions and learns a regressor from these embeddings.
result Better results than baseline instance-MIR across all datasets, state-of-the-art on two.
The reinforcement learning community has made great strides in designing algorithms capable of exceeding human performance on specific tasks. These algorithms are mostly trained one task at the time, each new task requiring to train a brand new agent instance. This means the learning algorithm is general, but each solu…
Generative models explain machine learning predictions with counterfactual instances.
problem Generating human-interpretable insights into machine learning model predictions.
method Sparse counterfactual explanations using conditional generative models.
result Single forward pass generates batches of counterfactual instances.
It is widely known in the machine learning community that class noise can be (and often is) detrimental to inducing a model of the data. Many current approaches use a single, often biased, measurement to determine if an instance is noisy. A biased measure may work well on certain data sets, but it can also be less effe…
A framework for reinforcement learning tackles CVRP with competitive results.
problem Optimizing routes for vehicles with limited capacity.
method Formulates action selection as a mixed-integer optimization problem, uses policy iteration to improve policies.
result Achieves an average gap of 1.7% with state-of-the-art OR methods on CVRP instances.
Improves federated learning robustness against data corruption.
problem Noise corruption of local agents in federated learning.
method Collaborative machine teaching with trusted instances.
result Effective identification and correction of training set bugs.
Adversarial self-play in two-player games has delivered impressive results when used with reinforcement learning algorithms that combine deep neural networks and tree search. Algorithms like AlphaZero and Expert Iteration learn tabula-rasa, producing highly informative training data on the fly. However, the self-play t…
Scalable approach for object pose estimation across domains.
problem Object pose estimation across different datasets and models.
method Multi-path learning: shared encoder, object-specific decoders.
result Generalizes well from synthetic to real data and across various instances.
A method to generate multi-label data from single positive annotations.
problem Generating multi-label datasets is costly and impractical.
method Single-to-multi-label (S2M) sampling using Markov chain Monte Carlo.
result S2M sampling enables high-quality multi-label data with minimal annotation cost.
Fraud detection is a difficult problem that can benefit from predictive modeling. However, the verification of a prediction is challenging; for a single insurance policy, the model only provides a prediction score. We present a case study where we reflect on different instance-level model explanation techniques to aid …
Dynamic neural network toolkits such as PyTorch, DyNet, and Chainer offer more flexibility for implementing models that cope with data of varying dimensions and structure, relative to toolkits that operate on statically declared computations (e.g., TensorFlow, CNTK, and Theano). However, existing toolkits - both static…
New tighter bounds for learning algorithms from Steinke & Zakynthinou's supersample setting.
problem Improving generalization bounds for machine learning algorithms.
method Information-theoretic approach using projected loss and Rademacher sequence.
result The new bounds are tighter than previous information-theoretic bounds.
ECG signal learning predicts cardiovascular death risk.
problem Scarce positive ECG event examples and class imbalance.
method Multiple instance learning framework for raw ECG signals.
result Learned risk score outperforms existing metrics.
Patient pain can be detected highly reliably from facial expressions using a set of facial muscle-based action units (AUs) defined by the Facial Action Coding System (FACS). A key characteristic of facial expression of pain is the simultaneous occurrence of pain-related AU combinations, whose automated detection would …
A deep learning framework separates overlapping nuclei in histology images.
problem Challenges in nuclear segmentation due to overlapping nuclei.
method Proposal-free spatially-aware deep learning framework with multi-scale spatial information.
result State-of-the-art performance in nuclear segmentation on a multi-organ data set.
LADaR framework calibrates machine learning models for instance-wise predictions.
problem Challenges in assessing and calibrating predictive distributions for complex inputs.
method Local Amortized Diagnostics and Reshaping of Conditional Densities (LADaR) framework and extttCal−PIT algorithm. result Achieves better instance-wise calibration than existing methods in galaxy distance estimation.
Predicting the completion time of business process instances would be a very helpful aid when managing processes under service level agreement constraints. The ability to know in advance the trend of running process instances would allow business managers to react in time, in order to prevent delays or undesirable situ…
There are classification tasks that take as inputs groups of images rather than single images. In order to address such situations, we introduce a nested multi-instance deep network. The approach is generic in that it is applicable to general data instances, not just images. The network has several convolutional neural…
After building a classifier with modern tools of machine learning we typically have a black box at hand that is able to predict well for unseen data. Thus, we get an answer to the question what is the most likely label of a given unseen data point. However, most methods will provide no answer why the model predicted th…
Deep learning matches classical feature-based AS models for TSP.
problem Automated selection of algorithms for the TSP.
method Evolved instances, deep neural network, visual representation.
result Deep learning approach matches classical feature-based models.
New concept of partial comonotonicity connects riskmetrics and dependence.
problem Understanding and quantifying risk metrics under partial comonotonicity.
method Developed a new notion of partial comonotonicity and established its connection to distortion riskmetrics.
result Partial comonotonicity uniquely characterizes a class of distortion riskmetrics through additivity.
We present an end-to-end framework for solving the Vehicle Routing Problem (VRP) using reinforcement learning. In this approach, we train a single model that finds near-optimal solutions for problem instances sampled from a given distribution, only by observing the reward signals and following feasibility rules. Our mo…
Convolutional neural networks (CNN) have achieved state of the art performance on both classification and segmentation tasks. Applying CNNs to microscopy images is challenging due to the lack of datasets labeled at the single cell level. We extend the application of CNNs to microscopy image classification and segmentat…
Instance embeddings are an efficient and versatile image representation that facilitates applications like recognition, verification, retrieval, and clustering. Many metric learning methods represent the input as a single point in the embedding space. Often the distance between points is used as a proxy for match confi…