This paper bridges offline and online RL by studying policy finetuning with a reference policy.
problem Sample-efficient reinforcement learning in online and offline settings.
method Design of policy finetuning algorithms and analysis of sample complexity.
result Theoretical analysis shows that the optimal policy finetuning algorithm is either offline reduction or purely online RL.
Algorithm improves reinforcement learning policies using offline data.
problem Improving reinforcement learning policies with limited online data.
method Designs a single non-reactive policy using offline data with provable guarantees.
result Algorithm achieves better policy quality with less online data.
Transforming sparse outcomes into dense process rewards for efficient reinforcement learning.
problem Training RL policies to maximize sparse outcomes.
method Incentivizing policy matching state-action visitations of successful episodes.
result Significantly faster RL finetuning performance.
Vanishing gradients hinder reinforcement finetuning of language models.
problem Vanishing gradients impede the optimization of language models using reinforcement finetuning.
method The study identifies vanishing gradients as a fundamental optimization obstacle in reinforcement finetuning and proposes an initial supervised finetuning phase to mitigate this issue.
result An initial supervised finetuning phase is crucial for successful reinforcement finetuning of language models, as it helps prevent vanishing gradients and maximizes rewards.
Improves drug properties using a novel LLM and reinforcement learning.
problem Optimizing drug properties while retaining chemical stability.
method Structured Policy Optimization (SPO) for fine-tuning a large language model.
result Enhanced drug properties across multiple target objectives.
SkMM selects data for finetuning by balancing bias and variance.
problem Balancing bias and variance in high-dimensional finetuning.
method Gradient sketching for bias reduction and moment matching for variance reduction.
result Gradient sketching selects samples efficiently and accurately.
Paper introduces a multi-stage influence function to track model predictions.
problem Improving natural language processing and computer vision performance.
method Develops a multi-stage influence function to track predictions from finetuned models back to pretraining data.
result Identifies pretraining examples contributing most to finetuning task predictions.
PLoP optimizes LoRA placement for efficient large model finetuning.
problem Improving efficiency of LoRA adaptation for large models.
method Intuitive theoretical analysis for automatic adapter placement.
result PLoP consistently outperforms existing placement strategies.
In natural language processing, it has been observed recently that generalization could be greatly improved by finetuning a large-scale language model pretrained on a large unlabeled corpus. Despite its recent success and wide adoption, finetuning a large pretrained language model on a downstream task is prone to degen…
Unified finetuning of all quantization degrees of freedom achieves state-of-the-art 4-bit quantization.
problem Achieving high accuracy in quantized neural networks while maintaining speed and resource constraints.
method Quantization-aware finetuning (QFT) that jointly optimizes all quantization degrees of freedom.
result 4-bit weight quantization results on-par with state-of-the-art (SoTA) within PTQ constraints.
Study shows pretraining and finetuning can effectively tackle covariate shift in linear regression.
problem Linear regression under covariate shift where source and target distributions differ but conditional distribution remains similar.
method Pretraining on source data and finetuning on target data using online SGD.
result Transfer learning with O(N2) source data is as effective as supervised learning with N target data. LoRAs enable efficient adaptation of large models; this paper explores processing LoRA weights with machine learning.
problem Efficient processing of low-rank weight decompositions in large finetuned models.
method Developed symmetry-aware invariant and equivariant LoL models to process LoRA weights.
result LoL models can predict CLIP scores, finetuning data attributes, and accuracy on downstream tasks.
We present relay policy learning, a method for imitation and reinforcement learning that can solve multi-stage, long-horizon robotic tasks. This general and universally-applicable, two-phase approach consists of an imitation learning stage that produces goal-conditioned hierarchical policies, and a reinforcement learni…
New framework reduces LLM complexity by directly finetuning in Boolean domain.
problem Reducing the complexity of large language models (LLMs) while maintaining performance.
method Proposes a novel framework using multi-kernel Boolean parameters for direct finetuning in the Boolean domain.
result Significantly reduces complexity during both finetuning and inference, outperforming recent techniques.
Meta-learning techniques like MAML and Reptile fail to generalize well to out-of-distribution tasks compared to simple finetuning.
problem Generalization of meta-learning techniques in low-data scenarios.
method Investigation of MAML, Reptile, and finetuning on various tasks.
result MAML and Reptile specialize for fast adaptation but fail to generalize to out-of-distribution tasks.
FinFlowRL learns from experts to optimize financial control in changing markets.
problem Traditional finance control methods fail in real-world, non-stationary markets.
method Imitation-Reinforcement Learning framework that pretrains on expert strategies and finetunes in noise space.
result Consistently outperforms individually optimized experts across diverse market conditions.
Study shows different initialization schemes for LoRA finetuning impact performance.
problem The impact of initialization schemes on LoRA finetuning performance.
method Compared two initialization schemes: B=0, A=random vs. A=0, B=random.
result First initialization scheme yields better performance on average.
A pretrained LLM and its finetuned version can detect OOD data effectively.
problem Detecting out-of-distribution data in language models.
method Using the likelihood ratio between a pretrained and finetuned LLM.
result The likelihood ratio is an effective criterion for OOD detection.
FinFlowRL combines imitation and reinforcement learning for better financial control.
problem Traditional stochastic control methods fail in real-world finance due to changing market conditions.
method FinFlowRL uses imitation learning to pretrain an adaptive meta policy, then finetunes it with reinforcement learning.
result FinFlowRL consistently outperforms individual strategies across various market conditions.
ConMeZO speeds up zeroth-order optimization for large language models.
problem Slow convergence in high-dimensional parameter spaces of large language models.
method Adaptive directional sampling in a cone centered around a momentum estimate.
result Achieves the same convergence rate as MeZO but up to 2X faster.
Vision transformers benefit from non-smooth components in adaptation.
problem Understanding the role of non-smoothness in vision transformer adaptation.
method Theoretical analysis and extensive experiments on large-scale vision transformers.
result High plasticity of attention modules and feedforward layers leads to better finetuning performance.
Efficiently adapting to new environments and changes in dynamics is critical for agents to successfully operate in the real world. Reinforcement learning (RL) based approaches typically rely on external reward feedback for adaptation. However, in many scenarios this reward signal might not be readily available for the …
Paper introduces RiskEmbed, a finetuned model for financial risk management.
problem Improving retrieval accuracy in financial question-answering systems.
method Curated dataset and finetuned BERT model for financial domain.
result RiskEmbed significantly outperforms general-purpose and financial embedding models.
BioFinBERT analyzes sentiment of biotech press releases and financial text around inflection points.
problem Analyzing sentiment of biotech press releases and financial text around inflection points.
method Finetuning BioBERT on financial datasets to create BioFinBERT for sentiment analysis.
result BioFinBERT accurately analyzes sentiment of biotech press releases and financial text around inflection points.
IVON improves LoRA finetuning with minimal overhead and significant accuracy gains.
problem Improving LoRA finetuning with Bayesian methods.
method IVON, a variational algorithm, with posterior pruning.
result Significant accuracy improvements over AdamW and other Bayesian methods.
BoRA finetunes multi-task LLMs by sharing information through hierarchical priors.
problem Limited data for some tasks in multi-task LLMs.
method Bayesian hierarchical low-rank adaption.
result BoRA outperforms individual and unified model approaches.
The paper proposes a learning algorithm that improves adaptability and generalization.
problem Improving adaptability and generalization in learning models.
method Learning to meta-learn by meta-finetuning on related tasks before adapting to specific tasks.
result Learning to meta-learn improves adaptability and generalization across various tasks.
New Bayesian method improves Pareto front estimation in multitask finetuning.
problem Efficiently estimating Pareto fronts for multitask finetuning.
method Variational Model Merging using non-Gaussian posteriors.
result More flexible posteriors lead to better Pareto front estimates.
Study reveals attributes help in few-shot classification generalization.
problem Understanding what makes some novel classes easier to learn.
method Defined attributes to quantify concept relatedness, used supervised and self-supervised learning.
result Combining self-supervised pretraining with supervised finetuning improves generalization.
Optimization-based pruning eliminates backpropagation for large language models.
problem Suboptimal pruning performance due to heuristic metrics.
method Optimization of Bernoulli distribution to learn pruning masks without backpropagation.
result Efficient pruning of large language models with improved performance.
WebGUM learns web navigation from multimodal data, outperforming previous methods.
problem Limited generalization from domain-specific models in web navigation.
method Instruction-following multimodal agent trained on vision-language foundation models.
result Significant improvement in web navigation performance on benchmarks.
Improves retrieval accuracy for hierarchical documents, especially for distant matches.
problem Limited expressive power of dual encoder models in hierarchical retrieval.
method Proves feasibility of DEs for HR, introduces pretrain-finetune recipe to improve long-distance retrieval.
result Pretrain-finetune boosts recall on long-distance pairs from 19% to 76%.
Study improves cryptocurrency price prediction using unlabeled text data.
problem Predicting cryptocurrency returns from unlabelled text data.
method Introduced weak learning approach to finetune BERT on unlabeled text data.
result Finetuning pretrained NLP models with weak labels enhances forecast accuracy.
Meta learning optimizes neural network quantization for efficient inference.
problem Uniform bitwidth quantization is sub-optimal for neural network compression.
method Meta learning to automatically generate hybrid quantization policies.
result Meta learning outperforms uniform quantization and RL approaches.
MPP trains a transformer to predict multiple physical systems, improving accuracy across various tasks.
problem Training models for specific physical systems is inefficient and requires fine-tuning.
method MPP trains a shared transformer on multiple heterogeneous physical systems, projecting fields into a shared embedding space.
result A single MPP-pretrained transformer outperforms task-specific models on all pretraining sub-tasks and downstream tasks.
This dissertation automates deep learning pipelines and uses meta-learning for better model selection and data augmentation.
problem Challenges in selecting and fine-tuning deep learning pipelines for new datasets.
method Meta-learning for DL pipeline selection and data augmentation, using synthetic data.
result Meta-learned approaches outperform traditional methods in automated DL pipeline selection and data augmentation.
Proposes a new algorithm for efficient online model selection of LLMs considering the increasing-then-converging trend.
problem Balancing cost and performance in choosing the best LLM among a diverse set of models.
method Introduces a time-increasing bandit algorithm (TI-UCB) that predicts model performance increases and balances exploration and exploitation.
result Achieves a logarithmic regret upper bound, indicating a fast convergence rate in model selection.
APQ jointly optimizes neural architecture, pruning, and quantization for efficient inference.
problem Efficient deep learning inference on resource-constrained hardware.
method Joint optimization of neural architecture, pruning, and quantization policy using a quantization-aware accuracy predictor.
result Joint optimization leads to 2.3% higher ImageNet accuracy with reduced latency and energy consumption.
CFA improves model's ability to generalize across unseen domain-class combinations.
problem Challenges in real-world machine learning applications due to data distribution shifts and limited training data.
method Developed Compositional Feature Alignment (CFA) technique to improve CG ability of pretrained models.
result CFA outperforms common finetuning techniques in compositional generalization.
The paper studies how search and distillation improve reasoning in large language models.
problem Improving reasoning capabilities of large language models.
method Viewing chain-of-thought generation as a metastable Markov process, proving benefits of search and distillation.
result Search protocol rewards sparse edges, reducing the expected number of steps to reach different clusters.
This study examines how the size and alignment of pretraining data affect the performance of large language models on downstream tasks.
problem Understanding how the size and alignment of pretraining data impact the performance of large language models on downstream tasks.
method Investigated the scaling behavior of large language models in a transfer learning setting, focusing on machine translation tasks.
result The size of the finetuning dataset and the distribution alignment between pretraining and downstream data significantly influence the scaling behavior of downstream performance.
Simplifies transfer learning with deep neural networks using ridge regression.
problem High computational cost of finetuning deep models for transfer learning.
method Leverage the low-rank property of deep neural networks' feature vectors in kernel ridge regression.
result Successful on supervised and semi-supervised transfer learning tasks.
Many sleep studies suffer from the problem of insufficient data to fully utilize deep neural networks as different labs use different recordings set ups, leading to the need of training automated algorithms on rather small databases, whereas large annotated databases are around but cannot be directly included into thes…
Improved LoRA+ adapts large models more efficiently.
problem Suboptimal feature learning in LoRA for large models.
method Different learning rates for LoRA adapter matrices A and B.
result Improves performance and speed by 1-2% and up to 2X, respectively.
Prior work in visual dialog has focused on training deep neural models on VisDial in isolation. Instead, we present an approach to leverage pretraining on related vision-language datasets before transferring to visual dialog. We adapt the recently proposed ViLBERT (Lu et al., 2019) model for multi-turn visually-grounde…
It is a long-standing challenge to enable an intelligent agent to learn in one environment and generalize to an unseen environment without further data collection and finetuning. In this paper, we consider a zero shot generalization problem setup that complies with biological intelligent agents' learning and generaliza…
Self-supervised learning improves ECG classification performance.
problem Label scarcity in clinical 12-lead ECG data.
method Adapted self-supervised methods to ECG domain, focusing on contrastive representations and latent forecasting.
result Contrastive predictive coding adaptation yields linear evaluation performance only 0.5% below supervised performance.
New study finds best language model architecture and pretraining objective for zero-shot tasks.
problem Evaluating which language model architectures and pretraining objectives best enable zero-shot generalization.
method Compared three model architectures and two pretraining objectives across 170 billion tokens, with and without finetuning.
result Causal decoder-only models trained on autoregressive language modeling exhibit strongest zero-shot generalization.