Proposes a new method for continual learning in neural networks.
problem Challenges in applying sequential Bayesian inference to neural networks.
method Sequential function-space variational inference.
result Neural networks trained with the proposed method achieve better predictive accuracy.
VAR-GPs solve continual learning by updating posteriors sequentially.
problem Catastrophic forgetting in sequential learning tasks.
method Sparse inducing point approximations and auto-regressive variational distribution.
result VAR-GPs prevent catastrophic forgetting and outperform baselines.
Proposes a framework for semi-supervised continual learning from sequentially arriving data.
problem Learning from data with changing task distribution over time, especially in domains with a mix of labeled and unlabeled data.
method Meta-Consolidation for Continual Semi-Supervised Learning (MCSSL) framework with a hypernetwork and semi-supervised auxiliary classifier.
result Significant improvements in continual semi-supervised learning setting.
AI learns to learn sequentially without forgetting.
problem Preventing catastrophic forgetting in machine learning models.
method Meta-learning a neuromodulatory activation-gating function to control selective activation in deep neural networks.
result State-of-the-art continual learning performance with 600 classes (9,000 updates).
The paper proves differentiability of evolution maps in Lie groups.
problem Differentiability of evolution maps in infinite-dimensional Lie groups.
method Showed sequential continuity (Mackey k-continuity) leading to differentiability.
result Differentiability of evolution maps in C k C^k C k -semiregular Lie groups. Continuous-time algorithms improve online learning performance.
problem Online learning with sequential data and minimizing overall regret.
method Extending discrete-time algorithms to continuous-time models for online linear optimization, adversarial bandit, and adversarial linear bandit.
result Optimal regret bounds are proven for continuous-time settings.
SFSVI uses Gaussian mixtures to approximate neural network outputs for continual learning.
problem Learning new tasks without forgetting old ones in neural networks.
method Sequential function-space variational inference with Gaussian mixture approximation.
result Gaussian mixture SFSVI outperforms other methods in continual learning.
Bayesian approach helps update deep models without forgetting past data.
problem Updating deep models sequentially without forgetting past data.
method Bayesian inference for continual learning.
result Bayesian approach enables updating model beliefs with new data.
This research proposes a CL model for RNNs to handle sequential data without forgetting.
problem Learning in dynamic environments without forgetting previous knowledge for sequential data.
method A Recurrent Neural Network (RNN) model with Elastic Weight Consolidation (EWC) for CL.
result The proposed model outperforms EWC and RNNs on CL benchmarks for sequential data.
Extends GP models for sequential data, scalable and robust.
problem Handling sequential input-output observations in multi-task settings.
method Variational inference with sparse approximations and recursive GP priors.
result Tractable continual learning with KL divergences and recursive reconstruction.
New insights into continual learning for deep models, showing convergence issues but local linear solutions.
problem Challenges in continual learning for homogeneous deep models.
method Sequential projections onto task margin sets, leveraging nonconvex projection theory.
result Local linear convergence under certain conditions for homogeneous deep networks.
New method optimizes decision-making in uncertain environments.
problem Optimal decision-making under partial observability.
method Nested sequential Monte Carlo algorithm for continuous POMDPs.
result Demonstrated effectiveness on continuous POMDP benchmarks.
Most artificial intelligence models have limiting ability to solve new tasks faster, without forgetting previously acquired knowledge. The recently emerging paradigm of continual learning aims to solve this issue, in which the model learns various tasks in a sequential fashion. In this work, a novel approach for contin…
Study on how task sequence properties affect continual learning algorithms.
problem Understanding how task sequence properties influence continual learning algorithms.
method Proposes a new procedure using task space modeling and correlation analysis.
result Error rates are correlated to a task sequence's total complexity but not to sequential heterogeneity.
A Kalman Optimiser improves continual learning in deep neural networks.
problem Learning new tasks sequentially without forgetting previously learned knowledge.
method Divides the neural network into long-term and short-term memory units.
result The method enables continual learning and adaptation without forgetting.
A method to improve sequential learning by keeping past data errors in check.
problem Challenges in sequential learning with Gaussian processes due to accumulating errors.
method Memory-based dual sparse variational Gaussian processes.
result Improves accuracy in inference and learning for various applications.
This work proposes a model to prevent forgetting in continual classification learning.
problem Preventing forgetting in continual classification learning.
method The approach builds on lifelong generative capabilities and derives a new variational bound.
result The model prevents catastrophic forgetting in continual classification learning.
Residual Continual Learning prevents forgetting in sequential tasks.
problem Preventing catastrophic forgetting in sequential learning of multiple tasks.
method ResCL reparameterizes network parameters by combining original and fine-tuned networks, keeping network size constant.
result ResCL achieves state-of-the-art performance in various continual learning scenarios.
The paper extends consistency results for sequential design strategies to vector-valued Gaussian processes.
problem Estimating excursion sets of vector-valued Gaussian processes.
method Clarifying the connection between continuous Gaussian processes and Gaussian measures in Banach spaces, extending concepts and properties from scalar-valued settings to vector-valued settings.
result Consistency results for sequential design strategies can be applied to vector-valued Gaussian processes.
A new sequential test for unnormalized densities.
problem Testing unnormalized densities with adaptive stopping.
method Sequential kernelized Stein discrepancy test, using non-uniform Stein kernels.
result Valid test with asymptotic lower bound for growth.
Paper analyzes convergence of continual learning with adaptive methods.
problem Preventing catastrophic forgetting in sequential learning tasks.
method Adaptive method for nonconvex continual learning (NCCL) adjusts step sizes of previous and current tasks.
result Proposed adaptive method achieves same convergence rate as SGD when catastrophic forgetting is suppressed.
The paper tackles finding optimal treatment sequences in continuous state spaces.
problem Finding counterfactually optimal action sequences in continuous state spaces.
method Formalizes the problem using finite horizon Markov decision processes and structural causal models. Develops a search method based on the A* algorithm.
result The method can find optimal action sequences in polynomial time under certain conditions.
We present a novel framework for kernel learning with sequential data of any kind, such as time series, sequences of graphs, or strings. Our approach is based on signature features which can be seen as an ordered variant of sample (cross-)moments; it allows to obtain a "sequentialized" version of any static kernel. The…
This paper tackles sequential distribution shifts in representation learning.
problem Learning meaningful representations in a sequence of distribution shifts.
method Nonlinear Independent Component Analysis (ICA) framework for continual causal representation learning.
result The method achieves performance comparable to joint training on multiple offline distributions and shows no benefit from the incoming new distribution on all latent variables.
Study on neural networks' performance in sequential task learning.
problem Understanding the performance of neural networks in sequential task learning.
method Theoretical analysis of generalization performance in continual learning using statistical mechanical analysis of kernel ridge-less regression.
result Characteristic transitions from positive to negative transfer observed in neural networks.
FMT model improves multimodal sequential learning across language, vision, and acoustic data.
problem Modeling spatio-temporal dynamics across multiple modalities.
method Factorized Multimodal Transformer (FMT) that models intramodal and intermodal dynamics in a factorized manner.
result FMT outperforms existing models on 3 datasets and 21 labels, setting new state of the art.
New method improves ABI for sequential data, reducing forgetting and improving accuracy.
problem Performance degradation of ABI under model misspecification and distribution shifts.
method Decouples simulation-based pre-training from unsupervised SC fine-tuning, using memory buffer and elastic weight consolidation.
result Significant mitigation of forgetting and improved posterior estimates compared to standard simulation-based training.
Robot learns new tasks sequentially without forgetting past ones.
problem Teaching a robot to solve tasks in a continual learning scenario.
method Policy distillation and sim2real transfer.
result Robot can solve all encountered tasks without forgetting past ones.
CRUs model irregular time series with continuous hidden states.
problem Handling irregular time intervals in sequential data.
method Continuous Recurrent Units (CRUs) that integrate hidden states via a linear stochastic differential equation.
result CRUs outperform methods based on neural ordinary differential equations in irregular time series interpolation.
We extend an offline evaluation method for continuous-armed bandits.
problem Evaluating policies in continuous-armed bandits with limited data.
method Extend existing offline evaluation method for continuous-armed bandits.
result Our method provides a consistent ranking of policies.
We propose and analyze sequential design methods for the problem of ranking several response surfaces. Namely, given L ≥ 2 L \ge 2 L ≥ 2 response surfaces over a continuous input space X \cal X X , the aim is to efficiently find the index of the minimal response across the entire X \cal X X . The response surfaces are not known and ha…
We use deep reinforcement learning to optimize experimental designs efficiently.
problem Optimizing sequential experimental designs with limited exploration and black-box models.
method Reduced the optimal design problem to an MDP and solved it with deep reinforcement learning.
result Our approach achieves state-of-the-art performance on both continuous and discrete design spaces.
DualSMC combines filtering and planning for continuous POMDPs.
problem Handling multi-modal state distributions and uncertainty in continuous POMDPs.
method DualSMC network that combines SMC for filtering and planning, with adversarial particle filter and uncertainty-dependent policy.
result DualSMC effectively handles complex observations and remains interpretable.
Diffusion models generate music sequences without autoregressive loops.
problem Generating music sequences from symbolic data using diffusion models.
method Parameterize discrete symbolic data in continuous latent space, train diffusion model, generate sequences through reverse process.
result Strong unconditional generation and post-hoc conditional infilling compared to autoregressive models.
Developments in deep generative models have allowed for tractable learning of high-dimensional data distributions. While the employed learning procedures typically assume that training data is drawn i.i.d. from the distribution of interest, it may be desirable to model distinct distributions which are observed sequenti…
Neural nets learn and forget tasks sequentially, showing promising scalability.
problem Learning and forgetting of multiple visual tasks in a sequential setting.
method Simulated sequential learning of ten related visual tasks.
result Neural nets show forward facilitation and backward interference, which are key phenomena.
This paper reviews methods for interpreting deep learning models with sequential data.
problem Limited interpretability of deep learning models in sequential data domains.
method Reviews and compares techniques for sequential interpretability.
result Current techniques have limitations and future research is needed.
Improves text clustering by incorporating sequential features and word embeddings.
problem Lack of sequential information and synonym handling in current text clustering methods.
method SiDPMM model that models documents as joint of bags of words, sequential features, and word embeddings.
result Significant improvement in performance and accurate inference of cluster numbers.
Paper defines benchmarks for learning new tasks sequentially.
problem Efficient evaluation of continual few-shot learning.
method Theoretical framework and flexible benchmarks.
result Introduction of SlimageNet64 for efficient evaluation.
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.
CRPS improves GP-based sequential design for chemical space.
problem Finding molecules with specific properties in synthetic chemistry.
method Threshold-weighted CRPS as acquisition function for GP models in sequential design.
result Improved performance in molecule research with CRPS-based strategies.
The study examines robust decision-making in volatile financial markets, finding action robustness is more impactful than uncertainty tolerance.
problem Sequential decision making in high-frequency markets under evolving uncertainty.
method Analyzes two dimensions of robustness: uncertainty tolerance and action robustness, using simulations and empirical evidence.
result Action robustness has a larger impact on profitability than uncertainty tolerance, and excessive robustness can reduce profitability in illiquid markets.
Paper proposes a method to learn multiple tasks without forgetting, maintaining model compactness.
problem Lifelong learning in deep learning models, especially forgetting of previous tasks.
method Combines deep model compression, critical weights selection, and progressive network expansion in an iterative manner.
result Incremental learning without forgetting, maintaining model compactness.
SurvSurf predicts first hitting times for intermittent events without monotonic violations.
problem Predicting first hitting times for intermittent events with monotonicity guarantees.
method Partially monotonic neural network for sequential events, incorporating unobserved events.
result SurvSurf outperforms existing models in MSE and IBS metrics.
Efficiently infers coupled hidden Markov models with noisy discrete observations.
problem Intractable inference for coupled continuous-time Markov chains with discrete observations.
method Latent Interacting Particle Systems, look-ahead functions, twisted Sequential Monte Carlo sampling.
result Demonstrated effectiveness on latent SIRS model and wildfire spread dynamics.
The paper studies the distance from calibration in sequential prediction, proving upper and lower bounds.
problem The challenge is to measure and minimize the deviation from perfect calibration in sequential binary prediction.
method The approach involves proving an O ( T ) O(\sqrt{T}) O ( T ) upper bound and an Ω ( T 1 / 3 ) Ω(T^{1/3}) Ω ( T 1/3 ) lower bound, using structural results and minimax arguments. result An O ( T ) O(\sqrt{T}) O ( T ) upper bound on the calibration distance is achieved, with an Ω ( T 1 / 3 ) Ω(T^{1/3}) Ω ( T 1/3 ) lower bound showing the inherent difficulty. A new method for continual learning using Gaussian processes.
problem Avoiding forgetting previous tasks in continual learning.
method Bayesian inference over function space, Gaussian process with inducing points.
result Strong results on benchmarks, unifying rehearsal and Bayesian perspectives.
Study evaluates CL methods in RNNs, highlighting differences from feedforward networks.
problem Preventing catastrophic forgetting in RNNs processing sequential data.
method Comprehensive evaluation of CL methods, including elastic weight consolidation and hypernetworks.
result Weight-importance methods perform similarly regardless of sequence length but require more stability for high working memory demands.