Emergent misalignment is influenced by training dynamics, model priors, and data.
problem Emergent misalignment in models
method Exploring training dynamics, model priors, and data
result Activation deltas before and after narrow fine-tuning correlate with their similarities when measured with the last prompt-token activations.
EVA adapts LoRA for faster, more efficient fine-tuning.
problem Fast and efficient fine-tuning of large models for specific tasks.
method EVA uses directions capturing most activation variance for initialization, maximizing gradient signal and reducing parameters.
result EVA achieves faster convergence and higher average scores across tasks, reducing parameters.
Neurons in higher cortical areas, such as the prefrontal cortex, are known to be tuned to a variety of sensory and motor variables. The resulting diversity of neural tuning often obscures the represented information. Here we introduce a novel dimensionality reduction technique, demixed principal component analysis (dPC…
We present analytical exploration of novel activation functions as consequence of integration of several ideas leading to implementation and subsequent use in habitability classification of exoplanets. Neural networks, although a powerful engine in supervised methods, often require expensive tuning efforts for optimize…
A new robust prefix-tuning framework improves model robustness against adversarial attacks.
problem Lack of robustness in prefix-tuning for adversarial attacks.
method Leveraging layerwise activations of pretrained models for additional prefix finetuning during the test phase.
result Framework substantially improves robustness over strong baselines while maintaining comparable accuracy on clean texts.
A common analytical problem in neuroscience is the interpretation of neural activity with respect to sensory input or behavioral output. This is typically achieved by regressing measured neural activity against known stimuli or behavioral variables to produce a "tuning function" for each neuron. Unfortunately, because …
Improved neural population modeling using shared features and ensemble detection.
problem Missing shared coding properties in neural latent variable models.
method Feature sharing across tuning curves and soft clustering of neurons.
result More interpretable and better-performing neural population models.
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…
Enhances molecular design models by fine-tuning uncertainty-guided VAEs.
problem Fine-tuning pre-trained generative models for specific molecular property optimization.
method Uncertainty-guided fine-tuning of variational autoencoders in an active learning setting.
result Uncertainty-guided fine-tuning improves model performance across multiple molecular properties.
Improves decentralized learning by teleporting active nodes for better convergence.
problem Decentralized learning's convergence rate degrades with large node numbers.
method Activates a subset of nodes, fetches parameters from previous active nodes, updates, and performs gossip averaging on a small topology.
result Teleportation completely alleviates convergence rate degradation with proper node activation.
The L1 regularization (Lasso) has proven to be a versatile tool to select relevant features and estimate the model coefficients simultaneously and has been widely used in many research areas such as genomes studies, finance, and biomedical imaging. Despite its popularity, it is very challenging to guarantee the feature…
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.
Enhances drug discovery models by understanding human language.
problem Low predictive quality of activity prediction models in drug discovery.
method Proposes a novel architecture with separate chemical and natural language input modules and a contrastive pre-training objective.
result Improves predictive performance on few-shot and zero-shot learning benchmarks.
Broad adoption of machine learning techniques has increased privacy concerns for models trained on sensitive data such as medical records. Existing techniques for training differentially private (DP) models give rigorous privacy guarantees, but applying these techniques to neural networks can severely degrade model per…
GRASP removes spurious correlations in fine-tuned models, improving task performance and reducing bias.
problem Fine-tuned models can latch onto spurious correlations, leading to bias and reduced generalization.
method GRASP identifies and removes spurious correlations from model weights without removing latent factors.
result GRASP significantly reduces bias and improves task performance in various fine-tuning tasks.
We design a new algorithm for batch active learning with deep neural network models. Our algorithm, Batch Active learning by Diverse Gradient Embeddings (BADGE), samples groups of points that are disparate and high-magnitude when represented in a hallucinated gradient space, a strategy designed to incorporate both pred…
This paper explains how batch normalization auto-tunes the regularization parameter based on data statistics.
problem Batch normalization accelerates deep learning training but the exact relationship to regularization is unclear.
method Theoretical analysis and empirical validation of batch normalization's role in auto-tuning the regularization parameter.
result Batch normalization auto-tunes the regularization parameter based on data statistics.
Deep convolutional neural networks have achieved great success in various applications. However, training an effective DNN model for a specific task is rather challenging because it requires a prior knowledge or experience to design the network architecture, repeated trial-and-error process to tune the parameters, and …
Many neural network architectures rely on the choice of the activation function for each hidden layer. Given the activation function, the neural network is trained over the bias and the weight parameters. The bias catches the center of the activation, and the weights capture the scale. Here we propose to train the netw…
Random feature maps improve forecasting of chaotic dynamical systems.
problem Forecasting chaotic dynamical systems with high accuracy.
method Data-driven random feature maps with tanh activation, skip connections, and localization.
result Effective forecasting skill for dynamical systems with dimensions up to 512.
Smartwatches are increasingly being used to recognize human daily life activities. These devices may employ different kind of machine learning (ML) solutions. One of such ML models is Gradient Boosting Machine (GBM) which has shown an excellent performance in the literature. The GBM can be trained on available data set…
Recognising human activities from streaming videos poses unique challenges to learning algorithms: predictive models need to be scalable, incrementally trainable, and must remain bounded in size even when the data stream is arbitrarily long. Furthermore, as parameter tuning is problematic in a streaming setting, suitab…
A new Universal Activation Function improves performance across various machine learning tasks.
problem Achieving near optimal performance in different machine learning tasks.
method Optimization algorithms evolve the UAF's parameters to match the optimal activation function for each task.
result The UAF converges to near optimal performance in classification, quantification, and reinforcement learning tasks.
Improved neural quantization reduces accuracy loss to less than 1% with 4-bit weights.
problem Reducing accuracy loss in neural quantization below 8-bits.
method Layer-wise calibration and integer programming to optimize bit-width allocation.
result Less than 1% accuracy degradation with 4-bit weights and activations.
A new method uses active learning to improve chemical simulation efficiency.
problem Efficiently estimating equilibrium-based chemical simulations.
method Sequential data-driven approach using Gaussian process uncertainty.
result Significantly reduced number of function evaluations.
SCoreBO improves Bayesian optimization by learning hyperparameters and self-correcting.
problem Efficient hyperparameter tuning for Gaussian process models in Bayesian optimization.
method Introduces SAL and SCoreBO, which prioritize hyperparameter learning and perform simultaneous optimization and learning.
result SCoreBO outperforms state-of-the-art methods on traditional benchmarks and atypical tasks.
The paper analyzes the statistical cost of tuning kernel hyperparameters in robust regression.
problem Finding the best interpolant from a class of kernels with unknown hyperparameters under adversarial noise.
method Finite-sample guarantees, subsampling guarantee for linear regression, ε-net argument for discretizing kernel parameterizations.
result Hyperparameter optimization increases sample complexity by just a logarithmic factor, compared to known parameters.
Feature representations from pre-trained deep neural networks have been known to exhibit excellent generalization and utility across a variety of related tasks. Fine-tuning is by far the simplest and most widely used approach that seeks to exploit and adapt these feature representations to novel tasks with limited data…
BatchTopK SAEs improve GPT-2 and Gemma activations with adjustable sparsity.
problem Interpreting language model activations using sparse autoencoders.
method Adapting the TopK constraint to the batch-level, allowing variable number of active latents per sample.
result BatchTopK SAEs consistently outperform TopK SAEs in reconstructing GPT-2 and Gemma activations.
This paper explains how to optimize prompts for model adaptation.
problem Understanding and optimizing prompt tuning for model adaptation.
method Bayesian view and meta-learning to explain prompt optimization.
result Optimal prompting can be studied formally as conditioning Bayesian predictors.
New metric captures individual neuron tuning across neural networks.
problem Need a metric that respects individual neuron tuning across different neural networks.
method Derived a 'soft' permutation-based metric using optimal transport theory.
result Metric avoids counter-intuitive outcomes and captures geometric insights.
Token-adaptive FFN design improves LLM expressivity.
problem Fixed activation functions limit FFN expressivity in LLMs.
method Mixture of Activations (MoA) and learnable activations (LA).
result MoA achieves lower terminal loss and better scaling than baselines.
DISTANA improves weather prediction by inferring hidden factors from temperature data.
problem Inferring hidden factors in spatiotemporal processes without supervision.
method Enhanced DISTANA architecture for spatiotemporal data, active tuning for latent state inference.
result DISTANA achieves more accurate predictions than other methods, inferring hidden factors from temperature data.
Algorithm optimally estimates linear dynamical systems with active input selection.
problem Estimating parameters of linear dynamical systems efficiently.
method Active learning with adaptive input selection.
result Finite time bound and asymptotic optimality proven.
K-fold Cross Validation is commonly used to evaluate classifiers and tune their hyperparameters. However, it assumes that data points are Independent and Identically Distributed (i.i.d.) so that samples used in the training and test sets can be selected randomly and uniformly. In Human Activity Recognition datasets, we…
Adaptive batch sizes improve active learning efficiency and flexibility.
problem Fixed batch sizes in active learning are inefficient due to dynamic cost-speed trade-offs.
method Probabilistic Numerics framework that adaptively changes batch sizes based on integration error and precision objectives.
result Significant enhancement in learning efficiency and flexibility across various applications.
Unified framework for active learning problems using information theory.
problem Combining level set estimation and Bayesian optimization.
method Information-theoretic criterion and acquisition function.
result Unified framework achieves state-of-the-art performance.
We show LLMs can be locally linear, enabling better control of activations.
problem Suboptimal control of LLM activations during generation.
method Model LLM inference as a linear dynamical system, compute feedback controllers using Jacobians, and adapt classical control theory.
result Robust, fine-grained control of LLM activations across models and tasks.
Paper proposes active learning for mining conflict dynamics from textual data.
problem Lack of detailed information on conflict dynamics in existing datasets.
method Active learning with a large, encoder-only language model.
result Performance similar to human coding with reduced human annotation.
Neural networks are known to be vulnerable to adversarial examples. Carefully chosen perturbations to real images, while imperceptible to humans, induce misclassification and threaten the reliability of deep learning systems in the wild. To guard against adversarial examples, we take inspiration from game theory and ca…
Active-GRPO improves molecular optimization by actively deciding when to imitate or self-improve.
problem Training robust and efficient molecular optimization models with large language models.
method Active-GRPO combines imitation and reinforcement learning, upgrading references and policies dynamically.
result Improves molecular optimization performance, achieving statistically significant gains.
Paper proposes efficient RLHF methods for LLMs using active queries.
problem Efficiently aligning LLMs with human preferences using RLHF.
method Formalizes RLHF as a dueling bandit problem, introduces APPO and ADPO algorithms.
result ADPO achieves similar performance to state-of-the-art methods with fewer queries.
LSQ+ improves quantization of neural nets with Swish activations, achieving state-of-the-art results.
problem Quantization of neural nets with Swish activations, especially negative activations, leads to significant performance loss.
method Introduces learnable scale and offset parameters for asymmetric quantization, and uses MSE-based initialization for quantization parameters.
result Significantly outperforms LSQ for low-bit quantization of neural nets with Swish activations, achieving up to 5.6% gain with W2A2 quantization of EfficientNet-B0.
The weight initialization and the activation function of deep neural networks have a crucial impact on the performance of the training procedure. An inappropriate selection can lead to the loss of information of the input during forward propagation and the exponential vanishing/exploding of gradients during back-propag…
In many real-world AD applications including computer security and fraud prevention, the anomaly detector must be configurable by the human analyst to minimize the effort on false positives. One important way to configure the detector is by providing true labels (nominal or anomaly) for a few instances. Recent work on …
Consider the multivariate nonparametric regression model. It is shown that estimators based on sparsely connected deep neural networks with ReLU activation function and properly chosen network architecture achieve the minimax rates of convergence (up to logn-factors) under a general composition assumption on the re…
DFRot improves LLMs by reducing outlier and massive activation effects.
problem Reducing outlier and massive activation effects in rotated LLMs.
method Weighted loss function and orthogonal Procrustes transforms for rotation matrix refinement.
result DFRot achieves dual free (Outlier-Free and Massive Activation-Free) with significant improvements in perplexity.
New algorithm improves Gaussian process hyperparameter tuning for large datasets.
problem Scalable hyperparameter tuning for Gaussian processes on large datasets.
method Estimates smoothness and length-scale parameters in Matern kernel using novel loss functions.
result Improved uncertainty quantification over traditional methods.