The paper develops a theory for iterative self-improvement of models, proving conditions for better performance with easy-to-hard curricula.
problem Lack of theoretical foundation for iterative self-improvement in practical settings.
method Modeling self-improvement as maximum-likelihood fine-tuning on reward-filtered distributions and proving finite-sample guarantees.
result Explicit feedback loop and conditions for better performance with easy-to-hard curricula.
Curriculum learning shows marginal benefits over random ordering on standard datasets.
problem Investigating the benefits of curriculum learning over random ordering in training neural networks.
method Experiments with curriculum, anti-curriculum, and random-curriculum orderings on various datasets and conditions.
result Curriculum learning only marginally improves performance on standard datasets and is outperformed by random ordering.
Reinforcement learning algorithms use correlations between policies and rewards to improve agent performance. But in dynamic or sparsely rewarding environments these correlations are often too small, or rewarding events are too infrequent to make learning feasible. Human education instead relies on curricula--the break…
Analyzes when and why curricula help neural networks learn.
problem Why and when does curriculum learning help neural networks?
method Statistical physics methods applied to a neural network model.
result Curricula can modestly speed up learning but do not provide generalization benefit.
This work adapts RDT for mental program construction, showing benefits and costs.
problem Applying RDT to mental programs with trade-offs between description length, error, and computational costs.
method Proposed a three-way trade-off and used simulations and partial information decomposition.
result Constructing a shared program library provides global benefits but is sensitive to curricula.
Overparameterisation can limit the benefits of curriculum learning in neural networks.
problem The ineffectiveness of curriculum learning in deep learning applications.
method Analytical study connecting curriculum learning and overparameterisation in an online learning setting for a 2-layer network in the XOR-like Gaussian Mixture problem.
result High degree of overparameterisation can limit the benefit from curricula.
Unified framework for multi-objective curriculum learning in robotics.
problem Improving sample efficiency and final performance in robotic policy learning.
method Unified automatic curriculum learning framework with multi-task hyper-net and flexible memory mechanism.
result Superior performance compared to state-of-the-art methods in robotic manipulation tasks.
SAMPLR optimizes for ground truth in aleatoric parameters to avoid curriculum-induced covariate shift.
problem Curriculum learning shifts training distribution, leading to suboptimal policies in aleatoric settings.
method SAMPLR optimizes ground-truth utility function, avoiding curriculum-induced covariate shift.
result SAMPLR preserves optimality under ground-truth distribution, promoting robustness across various environments.
The major challenge of learning from multi-label data has arisen from the overwhelming size of label space which makes this problem NP-hard. This problem can be alleviated by gradually involving easy to hard tags into the learning process. Besides, the utilization of a diversity maintenance approach avoids overfitting …
CausalWorld benchmarks robotic manipulation tasks with causal structure for transfer learning.
problem Challenges in transferring learned skills to new robotic manipulation environments.
method Proposes a simulation-based benchmark with a combinatorial family of tasks.
result Demonstrates the feasibility of tasks in the benchmark and provides baseline results.
Teaches uncertainty in ML through practical examples.
problem Lack of uncertainty teaching in ML curricula.
method Developed a curriculum and use cases.
result Motivates adoption of uncertainty concepts in AI courses.
Curriculum Learning - the idea of teaching by gradually exposing the learner to examples in a meaningful order, from easy to hard, has been investigated in the context of machine learning long ago. Although methods based on this concept have been empirically shown to improve performance of several learning algorithms, …
The ERI is a new index for measuring exam readiness.
problem Measuring exam readiness in a clear and actionable way.
method The ERI combines six signals derived from practice and mock tests, formalizing axioms for component maps and the composite.
result The ERI is a composite score interpretable and actionable for exam readiness.
Automatically generates a deep RL curriculum for faster and more stable learning.
problem How to automatically generate a curriculum for deep RL agents.
method Interprets curriculum generation as an inference problem, learning task distributions progressively.
result Curricula significantly improve learning performance across various environments and deep RL algorithms.
Curriculum learning in reinforcement learning is a training methodology that seeks to speed up learning of a difficult target task, by first training on a series of simpler tasks and transferring the knowledge acquired to the target task. Automatically choosing a sequence of such tasks (i.e. a curriculum) is an open pr…
This research formalizes inductive generalization and proposes a new learning paradigm called Inductive Learning.
problem Generalization from easy to hard tasks, especially out-of-domain generalization.
method Formalizes inductive generalization, introduces Inductive Learning, and outlines steps to adapt techniques for learning model successors.
result A new learning paradigm (Inductive Learning) that emphasizes induction and universal properties of learning and computation.
By simulating the easy-to-hard learning manners of humans/animals, the learning regimes called curriculum learning~(CL) and self-paced learning~(SPL) have been recently investigated and invoked broad interests. However, the intrinsic mechanism for analyzing why such learning regimes can work has not been comprehensivel…
Entropy-based model for hierarchical learning from multiscale data.
problem Learning from data with auxiliary information and multiscale target functions.
method Entropy-based hierarchical learning model with multiscale entropies.
result Entropy-based model yields stronger guarantees than uniform convergence bounds.
Estimating the relative importance of each sample in a training set has important practical and theoretical value, such as in importance sampling or curriculum learning. This kind of focus on individual samples invokes the concept of sample-wise learnability: How easy is it to correctly learn each sample (cf. PAC learn…
RLVR training dynamics reveal an implicit curriculum that shapes learning progression.
problem Understanding how RLVR overcomes the long-horizon barrier.
method Developed a theory of training dynamics for RLVR on transformers, using Fourier analysis on finite groups.
result Mixed-difficulty training naturally follows an implicit curriculum, shaping the learning progression from easy to hard.
Trading-R1 uses LLMs for financial trading, improving risk-adjusted returns.
problem Lack of interpretability and trust in AI for finance.
method Supervised fine-tuning and reinforcement learning with a curriculum.
result Improved risk-adjusted returns and lower drawdowns compared to other models.
RLVR learning dynamics naturally create an implicit curriculum for transformers.
problem How rewards based on final outcomes help overcome the long-horizon barrier in reasoning models.
method Developed a theory of training dynamics for RLVR on transformers, using Fourier analysis on finite groups.
result Mixed-difficulty training induces an implicit curriculum that shapes the learning progression from easy to hard.
Shaping in humans and animals has been shown to be a powerful tool for learning complex tasks as compared to learning in a randomized fashion. This makes the problem less complex and enables one to solve the easier sub task at hand first. Generating a curriculum for such guided learning involves subjecting the agent to…
APT-Gen generates tasks to help RL learn in hard problems.
problem Learning in hard exploration problems.
method APT-Gen uses a task generator to create tasks from a parameterized space, balancing performance and similarity to target tasks.
result APT-Gen outperforms baselines in grid world and robotic manipulation tasks.
DreamerV3 learns diverse tasks with a single configuration.
problem Generalizing reinforcement learning across multiple domains.
method World model and farsighted strategies.
result Outperforms specialized methods across 150 diverse tasks.
Diffusion models learn simple statistics before complex ones, revealing a sample complexity exponent.
problem Understanding the learning dynamics of diffusion models.
method Empirical observations and theoretical analysis of diffusion models and denoisers.
result Diffusion models learn simple statistics (pair-wise correlations) at linear sample complexity, while higher-order statistics (e.g., fourth cumulant) require cubic sample complexity.
Meta-learning strategy improves few-shot classification performance.
problem Few-shot classification with deep neural networks struggles when labeled samples are limited.
method Proposes an easy-to-hard expert meta-training strategy to arrange training tasks based on task hardness.
result Meta-learners achieve better results with the proposed expert training strategy.
This work proposes a student-teacher network for predicting hospital admission locations.
problem Accurate prediction of hospital admission locations to optimize resource allocation.
method Reinforcement learning approach where a teacher network selects data batches for a student network.
result The approach outperforms state-of-the-art methods on tabular data and image recognition.
Neural network optimizes learning sequence for reading words.
problem Children struggle with learning to read words due to inconsistent spelling-sound correspondences.
method Used a neural network to structure learning trials to optimize generalization accuracy.
result Significant improvement in generalization accuracy compared to random or frequency-based sequences.
sGPO optimizes RLVR training by balancing inference FLOPs and training efficiency.
problem RLVR training allocates rollout budget without considering query difficulty.
method sGPO trades inference FLOPs for reduced training FLOPs.
result sGPO matches or exceeds baseline performance while reducing training compute by a factor of three.
We introduce Mix&Match (M&M) - a training framework designed to facilitate rapid and effective learning in RL agents, especially those that would be too slow or too challenging to train otherwise. The key innovation is a procedure that allows us to automatically form a curriculum over agents. Through such a curriculum …
Software helps teach latent variable methods in multivariate data analytics.
problem Challenges in understanding multivariate data analytics for students.
method Interactive software for comparing latent variable methods.
result Builds intuition on choosing methods and interpreting coefficients.
U-Cast simplifies AI weather forecasting with a standard U-Net and efficient training.
problem Complex AI models limit accessibility and cost for weather forecasting.
method Simple U-Net backbone, deterministic pre-training, and probabilistic fine-tuning with Monte Carlo Dropout.
result U-Cast matches or exceeds state-of-the-art models in accuracy while reducing training and inference costs.
New social and economic activities massively exploit big data and machine learning algorithms to do inference on people's lives. Applications include automatic curricula evaluation, wage determination, and risk assessment for credits and loans. Recently, many governments and institutions have raised concerns about the …
eva method improves RL models by adaptively creating prompts, boosting performance.
problem Fixed prompt distribution limits scalability of RL models post-training.
method Evolving Alignment via Asymmetric Self-Play (eva) approach.
result Significant performance boost on benchmarks, e.g. 51.6% to 60.1% win-rate.
Teaches reproducible research to medical students and postgrads.
problem Lack of reproducibility in medical research practices.
method Designed and delivered a lecture series on reproducible research.
result Encountered practical obstacles in reproducing a published analysis.
This paper introduces CENIE to quantify environment novelty for better UED.
problem Challenges in measuring environment novelty for effective UED.
method CENIE framework using state-action space coverage and Gaussian Mixture Models.
result CENIE improves UED performance across multiple benchmarks.
New method learns optimal environment and goal difficulty for reinforcement learning.
problem Lack of robust transfer in reinforcement learning.
method Self-Supervised Active Domain Randomization (SS-ADR).
result Optimal domain randomization strategy improves transfer in goal-directed tasks.
Transformers can be made to implement specific algorithms by controlling training parameters.
problem Understanding when and how weight-tied looped transformers implement specific algorithms.
method Controlled experiments on group word problems, analyzing training contracts and convergence times.
result The speed and halting rule of weight-tied looped transformers are determined by the training contract and the number of loops.
Captures data influence changes during training.
problem Traditional influence functions fail for modern training methods.
method Formalized trajectory-specific LOO influence, using data value embedding.
result Data influence varies by training stage, early and late stages have greater impact.
Identifies learning rules from neural network observables.
problem Determine the underlying plasticity rules governing learning in biological systems.
method Simulated idealized neuroscience experiments with artificial neural networks to generate a dataset of learning trajectories. Used linear and non-linear classifiers to identify learning rules from aggregate statistics of weights, activations, and activity changes.
result Different classes of learning rules can be separated solely on the basis of aggregate statistics of the weights, activations, or instantaneous layer-wise activity changes.
AI could democratize education but risks inequality.
problem AI risks creating educational inequality.
method Speculation and posing questions on AI in education.
result AI could exacerbate educational inequality.
GTNs generate training data to accelerate AI learning.
problem Speeding up AI learning through better training data.
method Generative Teaching Networks (GTNs) learn to generate synthetic training data.
result GTNs accelerate AI learning and NAS evaluations.
A new method for solving complex sequential decision-making problems by decomposing them into multiple levels.
problem Sequential decision-making with natural multi-level structure.
method Multi-level meta-reinforcement learning with skill-based curriculum.
result Efficiently reduces stochasticity and policy search space, leading to fewer iterations and computations.