Offline optimization has recently emerged as an increasingly popular approach to mitigate the prohibitively expensive cost of online experimentation. The key idea is to learn a surrogate of the black-box function that underlines the target experiment using a static (offline) dataset of its previous input-output queries. Such an approach is, however, fraught with an out-of-distribution issue where the learned surrogate becomes inaccurate outside the offline data regimes. To mitigate this, existing offline optimizers have proposed numerous conditioning techniques to prevent the learned surrogate from being too erratic. Nonetheless, such conditioning strategies are often specific to particular surrogate or search models, which might not generalize to a different model choice. This motivates us to develop a model-agnostic approach instead, which incorporates a notion of model sharpness into the training loss of the surrogate as a regularizer. Our approach is supported by a new theoretical analysis demonstrating that reducing surrogate sharpness on the offline dataset provably reduces its generalized sharpness on unseen data. Our analysis extends existing theories from bounding generalized prediction loss (on unseen data) with loss sharpness to bounding the worst-case generalized surrogate sharpness with its empirical estimate on training data, providing a new perspective on sharpness regularization. Our extensive experimentation on a diverse range of optimization tasks also shows that reducing surrogate sharpness often leads to significant improvement, marking (up to) a noticeable 9.6% performance boost.

Model-X knockoff has garnered significant attention among various feature selection methods due to its guarantees for controlling the false discovery rate (FDR). Since its introduction in parametric design, knockoff techniques have evolved to handle arbitrary data distributions using deep learning-based generative models. However, we have observed limitations in the current implementations of the deep Model-X knockoff framework. Notably, the swap property that knockoffs require often faces challenges at the sample level, resulting in diminished selection power. To address these issues, we develop Deep Dependency Regularized Knockoff (DeepDRK), a distribution-free deep learning method that effectively balances FDR and power. In DeepDRK, we introduce a novel formulation of the knockoff model as a learning problem under multi-source adversarial attacks. By employing an innovative perturbation technique, we achieve lower FDR and higher power. Our model outperforms existing benchmarks across synthetic, semi-synthetic, and real-world datasets, particularly when sample sizes are small and data distributions are non-Gaussian.

Laplace approximation (LA) and its linearized variant (LLA) enable effortless adaptation of pretrained deep neural networks to Bayesian neural networks. The generalized Gauss-Newton (GGN) approximation is typically introduced to improve their tractability. However, LA and LLA are still confronted with non-trivial inefficiency issues and should rely on Kronecker-factored, diagonal, or even last-layer approximate GGN matrices in practical use. These approximations are likely to harm the fidelity of learning outcomes. To tackle this issue, inspired by the connections between LLA and neural target kernels (NTKs), we develop a Nystrom approximation to NTKs to accelerate LLA.

Large Transformer networks are increasingly used in settings where low inference latency is necessary to enable new applications and improve the end-user experience.However, autoregressive inference is resource intensive and requires parallelism for efficiency.Parallelism introduces collective communication that is both expensive and represents a phase when hardware resources are underutilized.Towards mitigating this, Kraken is an evolution of the standard Transformer architecture that is designed to complement existing tensor parallelism schemes for efficient inference on multi-device systems.By introducing a fixed degree of intra-layer model parallelism, the architecture allows collective operations to be overlapped with compute, decreasing latency and increasing hardware utilization.When trained on OpenWebText, Kraken models reach a similar perplexity as standard Transformers while also preserving their language modeling capabilities as evaluated on the SuperGLUE benchmark.Importantly, when tested on multi-GPU systems using TensorRT-LLM engines, Kraken speeds up Time To First Token by a mean of 35.6% across a range of model sizes, context lengths, and degrees of tensor parallelism.

Deep predictive models of neuronal activity have recently enabled several new discoveries about the selectivity and invariance of neurons in the visual cortex.These models learn a shared set of nonlinear basis functions, which are linearly combined via a learned weight vector to represent a neuron's function.Such weight vectors, which can be thought as embeddings of neuronal function, have been proposed to define functional cell types via unsupervised clustering.However, as deep models are usually highly overparameterized, the learning problem is unlikely to have a unique solution, which raises the question if such embeddings can be used in a meaningful way for downstream analysis.In this paper, we investigate how stable neuronal embeddings are with respect to changes in model architecture and initialization. We find that $L_1$ regularization to be an important ingredient for structured embeddings and develop an adaptive regularization that adjusts the strength of regularization per neuron. This regularization improves both predictive performance and how consistently neuronal embeddings cluster across model fits compared to uniform regularization.To overcome overparametrization, we propose an iterative feature pruning strategy which reduces the dimensionality of performance-optimized models by half without loss of performance and improves the consistency of neuronal embeddings with respect to clustering neurons.Our results suggest that to achieve an objective taxonomy of cell types or a compact representation of the functional landscape, we need novel architectures or learning techniques that improve identifiability.

Generative text-to-image models enable us to synthesize unlimited amounts of images in a controllable manner, spurring many recent efforts to train vision models with synthetic data. However, every synthetic image ultimately originates from the upstream data used to train the generator. Does the intermediate generator provide additional information over directly training on relevant parts of the upstream data? Grounding this question in the setting of image classification, we compare finetuning on task-relevant, targeted synthetic data generated by Stable Diffusion---a generative model trained on the LAION-2B dataset---against finetuning on targeted real images retrieved directly from LAION-2B. We show that while synthetic data can benefit some downstream tasks, it is universally matched or outperformed by real data from the simple retrieval baseline. Our analysis suggests that this underperformance is partially due to generator artifacts and inaccurate task-relevant visual details in the synthetic images. Overall, we argue that targeted retrieval is a critical baseline to consider when training with synthetic data---a baseline that current methods do not yet surpass.

Existing prompt-tuning methods have demonstrated impressive performances in continual learning (CL), by selecting and updating relevant prompts in the vision-transformer models. On the contrary, this paper aims to learn each task by tuning the prompts in the direction orthogonal to the subspace spanned by previous tasks' features, so as to ensure no interference on tasks that have been learned to overcome catastrophic forgetting in CL. However, different from the orthogonal projection in the traditional CNN architecture, the prompt gradient orthogonal projection in the ViT architecture shows completely different and greater challenges, i.e., 1) the high-order and non-linear self-attention operation; 2) the drift of prompt distribution brought by the LayerNorm in the transformer block. Theoretically, we have finally deduced two consistency conditions to achieve the prompt gradient orthogonal projection, which provide a theoretical guarantee of eliminating interference on previously learned knowledge via the self-attention mechanism in visual prompt tuning. In practice, an effective null-space-based approximation solution has been proposed to implement the prompt gradient orthogonal projection. Extensive experimental results demonstrate the effectiveness of anti-forgetting on four class-incremental benchmarks with diverse pre-trained baseline models, and our approach achieves superior performances to state-of-the-art methods.

Thousands of new scientific papers are published each month. Such information overload complicates researcher efforts to stay current with the state-of-the-art as well as to verify and correctly attribute claims. We pose the following research question: Given a text excerpt referencing a paper, could an LM act as a research assistant to correctly identify the referenced paper? We advance efforts to answer this question by building a benchmark that evaluates the abilities of LMs in citation attribution. Our benchmark, CiteME, consists of text excerpts from recent machine learning papers, each referencing a single other paper. CiteME use reveals a large gap between frontier LMs and human performance, with LMs achieving only 4.2-18.5% accuracy and humans 69.7%. We close this gap by introducing CiteAgent, an autonomous system built on the GPT-4o LM that can also search and read papers, which achieves an accuracy of 35.3% on CiteME. Overall, CiteME serves as a challenging testbed for open-ended claim attribution, driving the research community towards a future where any claim made by an LM can be automatically verified and discarded if found to be incorrect.

To understand neural network behavior, recent works quantitatively compare different networks' learned representations using canonical correlation analysis (CCA), centered kernel alignment (CKA), and other dissimilarity measures. Unfortunately, these widely used measures often disagree on fundamental observations, such as whether deep networks differing only in random initialization learn similar representations. These disagreements raise the question: which, if any, of these dissimilarity measures should we believe? We provide a framework to ground this question through a concrete test: measures should have \emph{sensitivity} to changes that affect functional behavior, and \emph{specificity} against changes that do not. We quantify this through a variety of functional behaviors including probing accuracy and robustness to distribution shift, and examine changes such as varying random initialization and deleting principal components. We find that current metrics exhibit different weaknesses, note that a classical baseline performs surprisingly well, and highlight settings where all metrics appear to fail, thus providing a challenge set for further improvement.

Solving mathematical problems requires advanced reasoning abilities and presents notable challenges for large language models. Previous works usually synthesize data from proprietary models to augment existing datasets, followed by instruction tuning to achieve top-tier results. However, our analysis of these datasets reveals severe biases towards easy queries, with frequent failures to generate any correct response for the most challenging queries.Hypothesizing that difficult queries are crucial to learning complex reasoning, we propose (), a method that allocates difficult queries more trials during the synthesis phase, enabling more extensive training on difficult samples.Utilizing, we have created new datasets for mathematical problem-solving that focus more on difficult queries and are substantially smaller than previous ones. Remarkably, our synthesis process solely relies on a 7B-sized open-weight model, without reliance on the commonly used proprietary GPT-4.We fine-tune various base models on our datasets ranging from 7B to 70B in size, resulting in a series of strong models called .In comprehensive in-domain and out-of-domain evaluation on 6 mathematical benchmarks, outperforms vanilla rejection tuning significantly, being superior or comparable to previous arts, despite using much smaller datasets and no proprietary models.