This work makes vision networks more interpretable by identifying key neurons.
problem Complex deep networks are hard to interpret, hindering safety-critical applications.
method Uses stochastic local competition and multimodal models to dissect and interpret neurons.
result The method generates understandable descriptions for a few active neurons, revealing network decision-making.
Active learning improves inspection systems by using weakly labeled data.
problem Rapidly updating machine vision inspection systems in evolving manufacturing processes.
method Developed a methodology for active learning from weakly labeled data, addressing covariate shift with domain-adversarial training.
result Demonstrated that active learning can accelerate the annotation process and reduce false positives.
Paper compresses ViTs models by 60% with minimal accuracy loss.
problem Memory constraints in ViTs models.
method Activation-aware low-rank tensor approximations.
result 60% reduction in model size with <1% accuracy loss.
ActNN reduces neural network training memory by 2-bit quantization.
problem Limited memory in training neural networks.
method Randomly quantizes activations to 2 bits, proving convergence and proposing mixed-precision strategies.
result Reduces activation memory footprint by 12x with negligible accuracy loss.
Robotic grasping system learns to target objects from a single image.
problem Robotic grasping in unstructured environments.
method Distributed reinforcement learning, active vision, synchronous SGD.
result System learns to grasp unseen objects without retraining.
Survey of complex-valued neural networks for improved performance.
problem Lack of complex-valued neural networks in machine learning frameworks.
method Literature review of CVNNs.
result Advantages of CVNNs over real-valued neural networks.
Plex improves model reliability across vision and language tasks.
problem Improving model reliability in diverse decision-making tasks involving uncertainty and adaptation.
method Developed ViT-Plex and T5-Plex pretrained model extensions to evaluate and improve reliability across 40 datasets.
result Plex greatly improves state-of-the-art across reliability tasks, simplifying evaluation and performance.
As Computer Vision moves from a passive analysis of pixels to active analysis of semantics, the breadth of information algorithms need to reason over has expanded significantly. One of the key challenges in this vein is the ability to identify the information required to make a decision, and select an action that will …
Mobile V-MoEs scale down ViTs for resource-constrained vision tasks.
problem Scaling down Vision Transformers for resource-constrained applications.
method Sparse Mixture-of-Experts (MoEs) applied to entire images, with a stable training procedure.
result Mobile V-MoEs achieve better performance-efficiency trade-offs than dense ViTs.
Activity recognition is the ability to identify and recognize the action or goals of the agent. The agent can be any object or entity that performs action that has end goals. The agents can be a single agent performing the action or group of agents performing the actions or having some interaction. Human activity recog…
New methods for graph completion improve image annotation and neuroimaging.
problem Sequential acquisition of partial measurements on graphs.
method Optimization model in Fourier domain, adaptive submodularity.
result Promising results on images and neuroimaging applications.
A new network controls task execution in a single vision system.
problem Executing multiple tasks accurately and efficiently in a single network.
method A top-down control network modifies main recognition network activations based on selected task, image content, and spatial location.
result Significantly better results on four datasets compared to state-of-the-art approaches.
Capsule networks improve on traditional neural networks by using vector activations.
problem Comparing capsule networks to traditional neural networks to validate their benefits.
method Deep visualization analysis, feature encoding across vector components, and instantiation parameter encoding.
result Capsule features encode information differently and provide benefits in computer vision applications.
AI monitors social distancing and masks at manufacturing plants.
problem Ensuring safety of workers during post-COVID production.
method Computer vision and AI techniques for social distancing and mask detection.
result Real-time alerts prevent violations of social distancing and mask-wearing.
Efficiently selects nearest neighbors for labeling to speed up active learning.
problem Intractable active learning and search for large-scale unlabeled data.
method Restricts candidate pool to nearest neighbors of labeled set.
result Achieved similar performance to global approach but reduced computational cost by up to 3 orders of magnitude.
Supervised machine learning based state-of-the-art computer vision techniques are in general data hungry and pose the challenges of not having adequate computing resources and of high costs involved in human labeling efforts. Training data subset selection and active learning techniques have been proposed as possible s…
Paper proposes ADC framework to reduce ViT SL training communication overhead.
problem Reducing communication overhead in ViT SL training.
method Two parallel compression strategies: class-agnostic merging and token discarding.
result Significantly reduces communication overhead without sacrificing accuracy.
This paper improves ensemble learning for vision tasks by encouraging diversity in predictions.
problem Generating effective ensembles of neural networks for multi-modal data.
method Explicitly optimize a diversity inducing adversarial loss for learning stochastic latent variables.
result Significant improvements in classification accuracy and out-of-distribution detection compared to baselines.
CovidSens uses social media to track COVID-19 spread.
problem Accurate and timely dissemination of COVID-19 information.
method Social sensing to analyze online user data.
result Real-time COVID-19 spread tracking system.
A new ensemble of Gaussian processes improves active learning efficiency.
problem Efficiently labeling data in high-cost domains like medical imaging.
method Adaptive weighted ensemble of Gaussian processes (EGP) for active learning.
result EGP-based approaches outperform single GP-based active learning methods.
KD2M unifies feature matching in neural networks.
problem Matching neural network activations for knowledge transfer.
method Formalizes distribution matching strategy.
result New theoretical results for KD.
Body-worn cameras are now commonly used for logging daily life, sports, and law enforcement activities, creating a large volume of archived footage. This paper studies the problem of classifying frames of footage according to the activity of the camera-wearer with an emphasis on application to real-world police body-wo…
We introduce a data-free quantization method for deep neural networks that does not require fine-tuning or hyperparameter selection. It achieves near-original model performance on common computer vision architectures and tasks. 8-bit fixed-point quantization is essential for efficient inference on modern deep learning …
This paper presents a self-supervised method for visual detection of the active speaker in a multi-person spoken interaction scenario. Active speaker detection is a fundamental prerequisite for any artificial cognitive system attempting to acquire language in social settings. The proposed method is intended to compleme…
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.
SAP corrects model for label noise by identifying and removing noisy samples.
problem Label corruption degrades model performance; acquiring perfect labels is costly.
method SAP uses SVD to identify and project model weights onto a clean activation space.
result SAP improves model generalization by up to 6% on CIFAR dataset with 25% synthetic corruption.
A fundamental problem in computer vision is boundary estimation, where the goal is to delineate the boundary of objects in an image. In this paper, we propose a method which jointly incorporates geometric and topological information within an image to simultaneously estimate boundaries for objects within images with mo…
Active learning (AL) is a learning paradigm where an active learner has to train a model (e.g., a classifier) which is in principal trained in a supervised way, but in AL it has to be done by means of a data set with initially unlabeled samples. To get labels for these samples, the active learner has to ask an oracle (…
Introspects convolutional speech recognition models using Gradient-adjusted Neuron Activation Profiles.
problem Lack of interpretability in deep learning ASR models.
method Gradient-adjusted Neuron Activation Profiles (GradNAPs) for feature and representation visualization.
result Gains insight into how data is processed in convolutional ASR models.
Rectified Linear Units (ReLUs) are among the most widely used activation function in a broad variety of tasks in vision. Recent theoretical results suggest that despite their excellent practical performance, in various cases, a substitution with basis expansions (e.g., polynomials) can yield significant benefits from b…
DORA analyzes deep neural networks' internal representations to detect spurious correlations.
problem Detecting spurious correlations in deep neural networks' internal representations.
method DORA uses Extreme-Activation (EA) distance measure to assess representation similarities.
result Identifies internal representations capable of detecting spurious correlations.
Single neuron with ADA learns XOR and outperforms other functions.
problem Classifying linearly non-separable data.
method Proposed a new artificial neuron with apical dendrite activation.
result ADA function achieves 100% accuracy on XOR and superior performance on benchmark datasets.
Typically a classifier trained on a given dataset (source domain) does not performs well if it is tested on data acquired in a different setting (target domain). This is the problem that domain adaptation (DA) tries to overcome and, while it is a well explored topic in computer vision, it is largely ignored in robotic …
This work interprets GELU and related activations via a first-order loss function.
problem Understanding and optimizing activation functions in neural networks.
method Complementary interpretation using the Gaussian first-order loss function.
result Calibrated or learned uniform-threshold gates are competitive and often outperform GELU, ReLU, and SiLU/Swish.
This research integrates attention into XAI frameworks for better model explanations.
problem Improving the interpretability of transformer models.
method Developed two novel explanation methods: Shapley value decomposition and token-level directional derivatives.
result Attention weights can be meaningfully incorporated into XAI frameworks, enhancing transformer explainability.
The splendid success of convolutional neural networks (CNNs) in computer vision is largely attributable to the availability of massive annotated datasets, such as ImageNet and Places. However, in medical imaging, it is challenging to create such large annotated datasets, as annotating medical images is not only tedious…
Transformers exhibit sparse activation maps, reducing computational load and improving robustness.
problem Sparse activation in Transformer models.
method Extensive experiments on various Transformer architectures and tasks.
result Sparsity in Transformers is a prevalent phenomenon, reducing FLOP count and improving model robustness.
Very deep CNNs achieve state-of-the-art results in both computer vision and speech recognition, but are difficult to train. The most popular way to train very deep CNNs is to use shortcut connections (SC) together with batch normalization (BN). Inspired by Self- Normalizing Neural Networks, we propose the self-normaliz…
New method accelerates CNNs for mobile devices by approximating tensors and quantizing weights.
problem Efficiently compress and accelerate CNNs for mobile devices.
method Low-rank tensor approximation in Tucker format combined with quantization of weights and activations.
result Our method significantly improves CNN performance on various classification tasks.
Improved model robustness against corruptions using online adaptation.
problem Machine vision models' vulnerability to image corruptions like blurring or compression artefacts.
method Using corrupted images' statistics for unsupervised online adaptation to improve robustness.
result ResNet-50 achieves 62.2% mCE on ImageNet-C with adaptation, improving from 76.7% without.
We describe a mechanism by which artificial neural networks can learn rapid adaptation - the ability to adapt on the fly, with little data, to new tasks - that we call conditionally shifted neurons. We apply this mechanism in the framework of metalearning, where the aim is to replicate some of the flexibility of human …
Dropout schedules can be optimized to significantly reduce model test loss.
problem Improving model performance in neural networks.
method Developed a mean-field theory of dropout at the edge of chaos, proposing front-loaded dropout schedules.
result Front-loaded dropout schedules reduce test loss by 18-35% over constant dropout.
Paper analyzes why deeper layers of ViTs perform worse on out-of-distribution tasks.
problem Performance degradation of intermediate layers in ViTs under distribution shift.
method Extensive linear probing experiments across various benchmarks and fine-grained analysis of transformer modules.
result Probing feedforward network activations yields best performance under significant distribution shift.
Graph Neural Networks improve 3D object detection in LiDAR point clouds.
problem Challenges in processing LiDAR data due to its 3D geometry and massive volume.
method Proposes a Graph Neural Network (GNN) based framework for 3D object detection.
result GNNs successfully identify objects in 3D LiDAR point clouds.
Subspace clustering is a growing field of unsupervised learning that has gained much popularity in the computer vision community. Applications can be found in areas such as motion segmentation and face clustering. It assumes that data originate from a union of subspaces, and clusters the data depending on the correspon…
Improved action recognition in live videos with hybrid FR-DL method.
problem High computational costs and lack of temporal information in conventional action recognition.
method Automated selection of representative frames, feature extraction, background subtraction, HOG, deep neural network, LSTM, Softmax-KNN classifier.
result Significant improvement in accuracy and speed compared to state-of-the-art methods.
A new algorithm for efficiently removing specific classes from a model without retraining.
problem Class forgetting in machine learning models.
method Estimating retain and forget spaces using SVD, removing shared information, and updating weights.
result Achieved up to 1.38% accuracy improvement on ImageNet dataset with minimal samples.
Proposes Π-Nets, polynomial neural networks, for improved representation power.
problem Improving representation power in deep learning models.
method Introduces Π-Nets, a new class of deep polynomial neural networks. result Demonstrates Π-Nets outperform standard DCNNs and achieve state-of-the-art results.