SMT improves robotic long-horizon tasks by embedding and utilizing past observations.
problem Long-horizon tasks in partially observable environments require effective long-term memory.
method Scene Memory Transformer (SMT) embeds and uses attention to exploit spatio-temporal dependencies.
result SMT outperforms existing policies in visual navigation tasks.
A neural scene representation framework enforcing 3D transformations.
problem Learning 3D scene representations from images without 3D supervision.
method Introducing a loss enforcing equivariance of the scene representation with 3D transformations.
result Real-time neural rendering with comparable results to models requiring minutes for inference.
The ability to look multiple times through a series of pose-adjusted glimpses is fundamental to human vision. This critical faculty allows us to understand highly complex visual scenes. Short term memory plays an integral role in aggregating the information obtained from these glimpses and informing our interpretation …
Transformer model estimates keywords for better audio captioning.
problem Indeterminacy in word selection for audio events/scenes.
method Transformer-based model with keyword estimation.
result Achieved state-of-the-art performance in AAC.
Efficient model for foggy scene understanding in vehicles.
problem Challenging scene understanding and segmentation under foggy conditions.
method Domain adaptation and illumination-invariant image transformation.
result Outperforms state-of-the-art models in foggy scene understanding.
New dataset tests mental rotation from single images, improving model understanding of 3D scenes.
problem Understanding how a scene looks from a different viewpoint using a single image.
method Created CLEVR-MRT dataset, explored neural architectures for volumetric scene representations.
result Demonstrated the effectiveness of volumetric representations in answering mental rotation questions.
Transformers can perform well with less long-range memory.
problem The need for deep long-range memory in Transformers for language modeling.
method Performed interventions to show performance can be achieved with fewer long-range memories and by limiting attention range.
result Comparable performance can be achieved with 6X fewer long-range memories and better performance with limited attention range.
Predicting movement of objects while the action of learning agent interacts with the dynamics of the scene still remains a key challenge in robotics. We propose a multi-layer Long Short Term Memory (LSTM) autoendocer network that predicts future frames for a robot navigating in a dynamic environment with moving obstacl…
We present a signal representation framework called the sparse manifold transform that combines key ideas from sparse coding, manifold learning, and slow feature analysis. It turns non-linear transformations in the primary sensory signal space into linear interpolations in a representational embedding space while maint…
Global Memory Augmentation (GMAT) improves Transformer performance on long documents.
problem Large memory requirements of Transformer pairwise dot-product attention for long sequences.
method Integrates a dense global memory of length M into sparse Transformer blocks.
result Significant improvement on various tasks, including synthetic tasks, masked language modeling, and reading comprehension.
Active-memory mechanisms can replace self-attention in Transformers, but optimal results often require both.
problem Replacing self-attention with active-memory mechanisms in Transformers.
method Evaluation of various active-memory mechanisms in a Transformer model.
result Active-memory mechanisms can achieve comparable results to self-attention for language modeling, but optimal results are often achieved by combining both mechanisms.
Image classification with deep neural networks is typically restricted to images of small dimensionality such as 224 x 244 in Resnet models [24]. This limitation excludes the 4000 x 3000 dimensional images that are taken by modern smartphone cameras and smart devices. In this work, we aim to mitigate the prohibitive in…
Wi-GATr learns to simulate wireless signals with high accuracy and speed.
problem Inaccurate wireless signal propagation models limit modern communication system design.
method Wi-GATr uses a Geometric Algebra Transformer to learn from scene primitives.
result Wi-GATr achieves more accurate predictions than existing methods.
In microsurgery, lasers have emerged as precise tools for bone ablation. A challenge is automatic control of laser bone ablation with 4D optical coherence tomography (OCT). OCT as high resolution imaging modality provides volumetric images of tissue and foresees information of bone position and orientation (pose) as we…
Transformers can store facts efficiently using associative memories.
problem Understanding how transformers store and recall factual information.
method Proved linear scaling of storage capacities for linear and MLP associative memories, introduced a synthetic task, and analyzed gradient flow.
result Shallow transformers can achieve near optimal storage capacity for factual recall tasks using associative memories.
LLMs can be tricked into recalling facts based on context clues.
problem Manipulation of LLMs' factual recall through context changes.
method Mathematical exploration of transformers' associative memory properties.
result Transformers use self-attention and value matrix for associative memory.
Paper tackles time series forecasting weaknesses in Transformers.
problem Transformer's locality-agnostic and memory-bottleneck issues in time series forecasting.
method Proposes convolutional self-attention and LogSparse Transformer to address locality and memory issues.
result Improves forecasting accuracy for fine-granularity, long-term dependency time series.
New analysis tightens memory capacity of Hopfield models using spherical codes.
problem Optimizing memory capacity in modern Hopfield models and Kernelized Hopfield Models.
method Connecting Hopfield models to spherical codes in information theory, establishing an optimal capacity bound and a sub-linear algorithm.
result First tight and optimal asymptotic memory capacity for modern Hopfield models, matching known lower bounds.
New model separates object attributes for better perceptual grouping.
problem Perceptual grouping of complex visual scenes.
method Spatial mixture models with learnable priors.
result Outperforms state-of-the-art methods in perceptual grouping.
SPARTAN learns sparse interaction graphs between objects in scenes.
problem Capturing structured interactions between entities in complex settings remains challenging.
method SPARTAN is a Transformer-based world model that learns sparse, context-dependent interaction graphs through sparsity regularization on attention patterns.
result SPARTAN achieves significantly improved few-shot adaptation to dynamics changes and robustness against distractors.
GENESIS generates and samples 3D scenes by capturing object interactions.
problem Lack of models that explicitly capture object interactions in scene generation.
method Object-centric latent variables, spatial GMM, amortized inference, autoregressive prior.
result First object-centric generative model of 3D visual scenes.
Compressive Transformer learns long-range sequences by compressing past memories.
problem Learning long-range sequences in language and speech models.
method Compressive Transformer compresses past memories for efficient long-range sequence learning.
result State-of-the-art performance on language and speech benchmarks.
Understanding how images of objects and scenes behave in response to specific ego-motions is a crucial aspect of proper visual development, yet existing visual learning methods are conspicuously disconnected from the physical source of their images. We propose to exploit proprioceptive motor signals to provide unsuperv…
Compact Recurrent Transformer (CRT) improves Transformer efficiency for long sequences.
problem Efficiently scaling Transformer architecture to long sequences with limited compute resources.
method Combines shallow Transformer models with recurrent neural networks and persistent memory.
result CRT achieves comparable or superior performance to full-length Transformers with shorter segments and reduced FLOPs.
Scene text magnifier enhances readability for visually impaired.
problem Helps visually impaired read natural scene text.
method Four CNN-based networks: character erasing, extraction, magnify, synthesis.
result Effective text magnification without background alteration.
New model separates objects in scenes, enabling novel arrangements and depth.
problem Lack of modular, compositional scene modeling in generative models.
method Ensemble of generative models (experts) compete for explaining different parts of a scene.
result Model generates scenes with novel object arrangement and depth ordering.
New model adds persistent memory to self-attention layers for improved performance.
problem Improving transformer performance by removing feed-forward layers.
method Augmenting self-attention layers with persistent memory vectors.
result The model outperforms standard transformers on language modeling benchmarks.
Generative Adversarial Networks (GANs) can produce images of remarkable complexity and realism but are generally structured to sample from a single latent source ignoring the explicit spatial interaction between multiple entities that could be present in a scene. Capturing such complex interactions between different ob…
Sparse Transformers reduce memory and time requirements for long sequence modeling.
problem Quadratic growth in memory and time with transformer sequence length.
method Sparse factorizations of attention matrix, deeper network training, attention matrix recomputation, fast attention kernels.
result Sparse Transformers can model sequences up to tens of thousands of timesteps with hundreds of layers.
ROOTS learns to represent and render 3D scenes with object-centric models.
problem Learning to represent and render 3D scenes with object-centric compositionality.
method Probabilistic generative model for learning object representations and scene rendering from partial observations.
result The model can infer 3D object representations and render scenes from arbitrary viewpoints.
The paper studies scaling laws for associative memory mechanisms.
problem Understanding and optimizing learning and memorization processes.
method High-dimensional matrices of outer products of embeddings, relating to transformer models. Derived scaling laws with sample and parameter sizes. Extensive numerical experiments.
result Precise scaling laws and statistical efficiency of estimators.
Despite enormous progress in object detection and classification, the problem of incorporating expected contextual relationships among object instances into modern recognition systems remains a key challenge. In this work we propose Information Pursuit, a Bayesian framework for scene parsing that combines prior models …
SPACE models complex scenes by decomposing objects and backgrounds.
problem Scalability and unsupervised object-oriented scene representation learning.
method Generative latent variable model combining spatial-attention and scene-mixture approaches.
result SPACE achieves factorized object representations and decomposes complex scenes.
NeRF-VAE generates 3D scenes with geometric structure from few images.
problem Generating 3D scenes from few images with geometric consistency.
method Combines NeRF and VAE, incorporating shared geometric structure.
result NeRF-VAE can infer and render geometrically-consistent scenes from unseen environments.
Improved acoustic scene classification with factorized CNN.
problem Acoustic scene classification in varying environments.
method Large-margin factorized CNN with triplet loss.
result Improved performance and better generalization on unseen data.
A new method for estimating joint value functions in multi-scene reinforcement learning.
problem High variance in samples for policy gradient computations in multi-scene environments.
method Sparse attention mechanism over multiple value function hypotheses to approximate the true joint value function.
result Significant improvements in reward scores and enhanced navigation efficiency across OpenAI ProcGen environments.
Generative models learn from unlabeled videos via object segmentation and scene modeling.
problem Learning generative models from unlabelled videos.
method Decomposed into three subtasks: motion segmentation, background and foreground modeling, and scene sampling.
result Approach allows learning models that generalize beyond occlusions and represent scenes in a modular fashion.
Adaptive beamforming collapses in highly non-stationary environments, but the Universal Switching Beamformer resolves this by dynamically adjusting memory length.
problem Adaptive beamforming performance degrades in highly non-stationary environments.
method Integrating sequential prediction into the beamforming architecture.
result The USB achieves agility and precision in tracking highly non-stationary scenes.
Extended LSTM improves language modeling performance.
problem Improving LSTM for large-scale language modeling.
method Exponential gating, modified memory structure, and residual stacking.
result xLSTM outperforms state-of-the-art models in performance and scaling.
Generative Multisensory Network learns 3D scene representations from multiple modalities.
problem Learning robust 3D scene representations from multiple sensory modalities.
method Amortized Product-of-Experts for efficient inference and cross-modal generation.
result The model can infer modality-invariant 3D scene representations efficiently from various sensory modalities.
Pix2Shape learns 3D scene representations from single images without supervision.
problem Learning 3D scene information from a single image without supervision.
method Pix2Shape uses an encoder, decoder, and critic network to generate 2.5D surfel-based reconstructions.
result Pix2Shape can generate complex 3D scenes from a single image, scaling with on-screen resolution.
Paper improves reinforcement learning in multi-scene tasks.
problem Reducing sample variance in multi-scene reinforcement learning.
method Sparse dynamic value estimation using Gaussian mixture models.
result Significant improvements in reward scores and navigation efficiency.
RICH models scenes as hierarchical tree to learn and generate complex compositions.
problem Learning compositional structures between parts and objects in natural scenes.
method RICH uses a latent scene graph to organize entities into a tree structure and employs a top-down inference approach.
result RICH learns and generates complex scene hierarchies from unlabeled data.
To address Task 5 in the Detection and Classification of Acoustic Scenes and Events (DCASE) 2018 challenge, in this paper, we propose an ensemble learning system. The proposed system consists of three different models, based on convolutional neural network and long short memory recurrent neural network. With extracted …
Improves reinforcement learning agent's scene-specific value function.
problem High variance in samples for policy gradient computations in multi-scene environments.
method Proposes dynamic value estimation (DVE) for multiple MDPs, clustering value functions across scenes.
result Lower sample variance and more accurate scene-specific value function estimates.
Recently, multiple formulations of vision problems as probabilistic inversions of generative models based on computer graphics have been proposed. However, applications to 3D perception from natural images have focused on low-dimensional latent scenes, due to challenges in both modeling and inference. Accounting for th…
The wavelet scattering transform is an invariant signal representation suitable for many signal processing and machine learning applications. We present the Kymatio software package, an easy-to-use, high-performance Python implementation of the scattering transform in 1D, 2D, and 3D that is compatible with modern deep …
A novel method for visual question answering using scene graphs and reinforcement learning.
problem Answering free-form questions about images with deep linguistic and visual understanding.
method Context-driven, sequential reasoning based on scene graphs and reinforcement learning.
result Our method almost reaches human performance on the GQA dataset.