ICML 2024 Papers

New ML models are often derived from existing ones (e.g., through fine-tuning, quantization or distillation), forming an ecosystem where models are *related* to each other and can share structure or even parameter values. Managing such a large and evolving ecosystem of model derivatives is challenging. For instance, the overhead of storing all such models is high, and models may inherit bugs from related models, complicating error attribution and debugging. In this paper, we propose a model versioning and management system called MGit that makes it easier to store, test, update, and collaborate on related models. MGit introduces a lineage graph that records the relationships between models, optimizations to efficiently store model parameters, and abstractions over this lineage graph that facilitate model testing, updating and collaboration. We find that MGit works well in practice: MGit is able to reduce model storage footprint by up to 7$\times$. Additionally, in a user study with 20 ML practitioners, users complete a model updating task 3$\times$ faster on average with MGit.

Modern Hopfield networks have enjoyed recent interest due to their connection to attention in transformers. Our paper provides a unified framework for sparse Hopfield networks by establishing a link with Fenchel-Young losses. The result is a new family of Hopfield-Fenchel-Young energies whose update rules are end-to-end differentiable sparse transformations. We reveal a connection between loss margins, sparsity, and exact memory retrieval. We further extend this framework to structured Hopfield networks via the SparseMAP transformation, which can retrieve pattern associations instead of a single pattern. Experiments on multiple instance learning and text rationalization demonstrate the usefulness of our approach.

We propose HyperAgent, a reinforcement learning (RL) algorithm based on the hypermodel framework for exploration in RL. HyperAgent allows for the efficient incremental approximation of posteriors associated with an optimal action-value function ($Q^\star$) without the need for conjugacy and follows the greedy policies w.r.t. these approximate posterior samples. We demonstrate that HyperAgent offers robust performance in large-scale deep RL benchmarks. It can solve Deep Sea hard exploration problems with episodes that optimally scale with problem size and exhibits significant efficiency gains in the Atari suite. Implementing HyperAgent requires minimal code addition to well-established deep RL frameworks like DQN. We theoretically prove that, under tabular assumptions, HyperAgent achieves logarithmic per-step computational complexity while attaining sublinear regret, matching the best known randomized tabular RL algorithm.

To address the shortcomings of real-world datasets, robust learning algorithms have been designed to overcome arbitrary and indiscriminate data corruption. However, practical processes of gathering data may lead to patterns of data corruption that are localized to specific partitions of the training dataset. Motivated by critical applications where the learned model is deployed to make predictions about people from a rich collection of overlapping subpopulations, we initiate the study of *multigroup robust* algorithms whose robustness guarantees for each subpopulation only degrade with the amount of data corruption *inside* that subpopulation. When the data corruption is not distributed uniformly over subpopulations, our algorithms provide more meaningful robustness guarantees than standard guarantees that are oblivious to how the data corruption and the affected subpopulations are related. Our techniques establish a new connection between multigroup fairness and robustness.

Topological deep learning (TDL) is a rapidly evolving field that uses topological features to understand and design deep learning models. This paper posits that TDL is the new frontier for relational learning. TDL may complement graph representation learning and geometric deep learning by incorporating topological concepts, and can thus provide a natural choice for various machine learning settings. To this end, this paper discusses open problems in TDL, ranging from practical benefits to theoretical foundations. For each problem, it outlines potential solutions and future research opportunities. At the same time, this paper serves as an invitation to the scientific community to actively participate in TDL research to unlock the potential of this emerging field.

Vector quantization is a fundamental operation for data compression and vector search. To obtain high accuracy, multi-codebook methods represent each vector using codewords across several codebooks. Residual quantization (RQ) is one such method, which iteratively quantizes the error of the previous step. While the error distribution is dependent on previously-selected codewords, this dependency is not accounted for in conventional RQ as it uses a fixed codebook per quantization step. In this paper, we propose QINCo, a neural RQ variant that constructs specialized codebooks per step that depend on the approximation of the vector from previous steps. Experiments show that QINCo outperforms state-of-the-art methods by a large margin on several datasets and code sizes. For example, QINCo achieves better nearest-neighbor search accuracy using 12-byte codes than the state-of-the-art UNQ using 16 bytes on the BigANN1M and Deep1M datasets.

When the training dataset comprises a 1:1 proportion of dogs to cats, a generative model that produces 1:1 dogs and cats better resembles the training species distribution than another model with 3:1 dogs and cats. Can we capture this phenomenon using existing metrics? Unfortunately, we cannot, because these metrics do not provide any interpretability beyond “diversity". In this context, we propose a new evaluation protocol that measures the divergence of a set of generated images from the training set regarding the distribution of attribute strengths as follows. Singleattribute Divergence (SaD) reveals the attributes that are generated excessively or insufficiently by measuring the divergence of PDFs of individual attributes. Paired-attribute Divergence (PaD) reveals such pairs of attributes by measuring the divergence of joint PDFs of pairs of attributes. For measuring the attribute strengths of an image, we propose Heterogeneous CLIPScore (HCS) which measures the cosine similarity between image and text vectors with heterogeneous initial points. With SaD and PaD, we reveal the following about existing generative models. ProjectedGAN generates implausible attribute relationships such as baby with beard even though it has competitive scores of existing metrics. Diffusion models struggle to capture diverse colors in the datasets. The larger sampling timesteps of the latent diffusion model generate the more minor objects including earrings and necklace. Stable Diffusion v1.5 better captures the attributes than v2.1. Our metrics lay a foundation for explainable evaluations of generative models.

Recent years have witnessed a paradigm shift in deep learning from task-centric model design to task-agnostic representation learning and task-specific fine-tuning. Pretrained model representations are commonly evaluated extensively across various real-world tasks and used as a foundation for different downstream tasks. This paper proposes a solution for assessing the quality of representations in a task-agnostic way. To circumvent the need for real-world data in evaluation, we explore the use of synthetic binary classification tasks with Gaussian mixtures to probe pretrained models and compare the robustness-accuracy performance on pretrained representations with an idealized reference. Our approach offers a holistic evaluation, revealing intrinsic model capabilities and reducing the dependency on real-life data for model evaluation. Evaluated with various pretrained image models, the experimental results confirm that our task-agnostic evaluation correlates with actual linear probing performance on downstream tasks and can also guide parameter choice in robust linear probing to achieve a better robustness-accuracy trade-off.

Blind Image Quality Assessment (BIQA) mirrors subjective made by human observers. Generally, humans favor comparing relative qualities over predicting absolute qualities directly. However, current BIQA models focus on mining the "local" context, i.e., the relationship between information among individual images and the absolute quality of the image, ignoring the "global" context of the relative quality contrast among different images in the training data. In this paper, we present the Perceptual Context and Sensitivity BIQA (CSIQA), a novel contrastive learning paradigm that seamlessly integrates "global'' and "local'' perspectives into the BIQA. Specifically, the CSIQA comprises two primary components: 1) A Quality Context Contrastive Learning module, which is equipped with different contrastive learning strategies to effectively capture potential quality correlations in the global context of the dataset. 2) A Quality-aware Mask Attention Module, which employs the random mask to ensure the consistency with visual local sensitivity, thereby improving the model's perception of local distortions. Extensive experiments on eight standard BIQA datasets demonstrate the superior performance to the state-of-the-art BIQA methods.

This study investigates how weight decay affects the update behavior of individual neurons in deep neural networks through a combination of applied analysis and experimentation. Weight decay can cause the expected magnitude and angular updates of a neuron's weight vector to converge to a steady state we call rotational equilibrium. These states can be highly homogeneous, effectively balancing the average rotation---a proxy for the effective learning rate---across different layers and neurons. Our work analyzes these dynamics across optimizers like Adam, Lion, and SGD with momentum, offering a new simple perspective on training that elucidates the efficacy of widely used but poorly understood methods in deep learning. We demonstrate how balanced rotation plays a key role in the effectiveness of normalization like Weight Standardization, as well as that of AdamW over Adam with L2-regularization. Finally, we show that explicitly controlling the rotation provides the benefits of weight decay while substantially reducing the need for learning rate warmup.

In this work, we investigate how people use text-to-image models to generate desired target images. To study this interaction, we created ArtWhisperer, an online game where users are given a target image and are tasked with iteratively finding a prompt that creates a similar-looking image as the target. Through this game, we recorded over 50,000 human-AI interactions; each interaction corresponds to one text prompt created by a user and the corresponding generated image. The majority of these are repeated interactions where a user iterates to find the best prompt for their target image, making this a unique sequential dataset for studying human-AI collaborations. In an initial analysis of this dataset, we identify several characteristics of prompt interactions and user strategies. People submit diverse prompts and are able to discover a variety of text descriptions that generate similar images. Interestingly, prompt diversity does not decrease as users find better prompts. We further propose a new metric to quantify AI model *steerability* using our dataset. We define steerability as the expected number of interactions required to adequately complete a task. We estimate this value by fitting a Markov chain for each target task and calculating the expected time to reach an adequate score. We quantify and compare AI steerability across different types of target images and two different models, finding that images of cities and nature are more steerable than artistic and fantasy images. We also evaluate popular vision-language models to assess their image understanding and ability to incorporate feedback. These findings provide insights into human-AI interaction behavior, present a concrete method of assessing AI steerability, and demonstrate the general utility of the ArtWhisperer dataset.

Estimating the ratio of two probability densities from finitely many samples, is a central task in machine learning and statistics. In this work, we show that a large class of kernel methods for density ratio estimation suffers from error saturation, which prevents algorithms from achieving fast error convergence rates on highly regular learning problems. To resolve saturation, we introduce iterated regularization in density ratio estimation to achieve fast error rates. Our methods outperform its non-iteratively regularized versions on benchmarks for density ratio estimation as well as on large-scale evaluations for importance-weighted ensembling of deep unsupervised domain adaptation models.

Mixture-of-Experts (MoE) has gained increasing popularity as a promising framework for scaling up large language models (LLMs). However, the reliability assessment of MoE lags behind its surging applications. Moreover, when transferred to new domains such as in fine-tuning MoE models sometimes underperform their dense counterparts. Motivated by the research gap and counter-intuitive phenomenon, we propose $\texttt{MoE-RBench}$, the first comprehensive assessment of SMoE reliability from three aspects: $\textit{(i)}$ safety and hallucination, $\textit{(ii)}$ resilience to adversarial attacks, and $\textit{(iii)}$ out-of-distribution robustness. Extensive models and datasets are tested to compare the MoE to dense networks from these reliability dimensions. Our empirical observations suggest that with appropriate hyperparameters, training recipes, and inference techniques, we can build the MoE model more reliably than the dense LLM. In particular, we find that the robustness of SMoE is sensitive to the basic training settings. We hope that this study can provide deeper insights into how to adapt the pre-trained MoE model to other tasks with higher-generation security, quality, and stability. Codes are available at https://github.com/UNITES-Lab/MoE-RBench.

Learning representations that generalize under distribution shifts is critical for building robust machine learning models. However, despite significant efforts in recent years, algorithmic advances in this direction have been limited. In this work, we seek to understand the fundamental difficulty of out-of-distribution generalization with deep neural networks. We first empirically show that perhaps surprisingly, even allowing a neural network to explicitly fit the representations obtained from a teacher network that can generalize out-of-distribution is insufficient for the generalization of the student network. Then, by a theoretical study of two-layer ReLU networks optimized by stochastic gradient descent (SGD) under a structured feature model, we identify a fundamental yet unexplored feature learning proclivity of neural networks, feature contamination: neural networks can learn uncorrelated features together with predictive features, resulting in generalization failure under distribution shifts. Notably, this mechanism essentially differs from the prevailing narrative in the literature that attributes the generalization failure to spurious correlations. Overall, our results offer new insights into the non-linear feature learning dynamics of neural networks and highlight the necessity of considering inductive biases in out-of-distribution generalization.

With tremendous efforts on developing effective e-commerce models, conventional e-commerce models show limited success in generalist e-commerce modeling, and suffer from unsatisfactory performance on new users and new products – a typical out-of-domain generalization challenge. Meanwhile, large language models (LLMs) demonstrate outstanding performance in generalist modeling and out-of-domain generalizability in many fields. Toward fully unleashing their power for e-commerce, in this paper, we construct ECInstruct, the first open-sourced, large-scale, and high-quality benchmark instruction dataset for e-commerce. Leveraging ECInstruct, we develop eCeLLM, a series of e-commerce LLMs, by instruction-tuning general-purpose LLMs. Our comprehensive experiments and evaluation demonstrate that eCeLLM models substantially outperform baseline models, including the most advanced GPT-4, and the state-of-the-art task-specific models in in-domain evaluation. Moreover, eCeLLM exhibits excellent generalizability to out-of-domain settings, including unseen products and unseen instructions, highlighting its superiority as a generalist e-commerce model. Both the ECInstruct dataset and the eCeLLM models show great potential in empowering versatile and effective LLMs for e-commerce. ECInstruct and eCeLLM models are publicly accessible through this link.

*How can we detect if copyrighted content was used in the training process of a language model, considering that the training data is typically undisclosed?* We are motivated by the premise that a language model is likely to identify verbatim excerpts from its training text. We propose DE-COP, a method to determine whether a piece of copyrighted content is included in training. DE-COP's core approach is to probe an LLM with multiple-choice questions, whose options include both verbatim text and their paraphrases. We construct BookTection, a benchmark with excerpts from 165 books published prior and subsequent to a model's training cutoff, along with their paraphrases. Our experiments show that DE-COP outperforms the prior best method by 8.6% in detection accuracy (AUC) on models with logits available. Moreover, DE-COP also achieves an average accuracy of 72% for detecting suspect books on fully black-box models where prior methods give approximately 0% accuracy. The code and datasets are available at https://github.com/LeiLiLab/DE-COP.

Spontaneous neural activity, crucial in memory, learning, and spatial navigation, often manifests itself as repetitive spatiotemporal patterns. Despite their importance, analyzing these patterns in large neural recordings remains challenging due to a lack of efficient and scalable detection methods. Addressing this gap, we introduce *convSeq*, an unsupervised method that employs backpropagation for optimizing spatiotemporal filters that effectively identify these neural patterns. Our method’s performance is validated on various synthetic data and real neural recordings, revealing spike sequences with unprecedented scalability and efficiency. Significantly surpassing existing methods in speed, *convSeq* sets a new standard for analyzing spontaneous neural activity, potentially advancing our understanding of information processing in neural circuits.

We consider the problem of estimating the learning rate in adaptive methods, such as AdaGrad and Adam. We propose Prodigy, an algorithm that provably estimates the distance to the solution $D$, which is needed to set the learning rate optimally. At its core, Prodigy is a modification of the D-Adaptation method for learning-rate-free learning. It improves upon the convergence rate of D-Adaptation by a factor of $\mathcal{O}(\sqrt{\log(D/d_0)})$, where $d_0$ is the initial estimate of $D$. We test Prodigy on 12 common logistic-regression benchmark datasets, VGG11 and ResNet-50 training on CIFAR10, ViT training on Imagenet, LSTM training on IWSLT14, DLRM training on Criteo dataset, VarNet on Knee MRI dataset, as well as RoBERTa and GPT transformer training on BookWiki. Our experimental results show that our approach consistently outperforms D-Adaptation and reaches test accuracy values close to that of hand-tuned Adam.

Sharpness-Aware Minimization (SAM), which performs gradient descent on adversarially perturbed weights, can improve generalization by identifying flatter minima. However, recent studies have shown that SAM may suffer from convergence instability and oscillate around saddle points, resulting in slow convergence and inferior performance. To address this problem, we propose the use of a lookahead mechanism to gather more information about the landscape by looking further ahead, and thus find a better trajectory to converge. By examining the nature of SAM, we simplify the extrapolation procedure, resulting in a more efficient algorithm. Theoretical results show that the proposed method converges to a stationary point and is less prone to saddle points. Experiments on standard benchmark datasets also verify that the proposed method outperforms the SOTAs, and converge more effectively to flat minima.