论文检索

Deep Learning · Representation Learning

Ameya Prabhu, Shiven Sinha, Ponnurangam Kumaraguru, Philip Torr, Ozan Sener, Puneet Dokania

Continual learning has primarily focused on the issue of catastrophic forgetting and the associated stability-plasticity tradeoffs. However, little attention has been paid to the efficacy of continually learned representations, as representations are learned alongside classifiers throughout the learning process. Our primary contribution is empirically demonstrating that existing online continually trained deep networks produce inferior representations compared to a simple pre-defined random transforms. Our approach embeds raw pixels using a fixed random transform, approximating an RBF-Kernel initialized before any data is seen. We then train a simple linear classifier on top without storing any exemplars, processing one sample at a time in an online continual learning setting. This method, called RanDumb, significantly outperforms state-of-the-art continually learned representations across all standard online continual learning benchmarks. Our study reveals the significant limitations of representation learning, particularly in low-exemplar and online continual learning scenarios. Extending our investigation to popular exemplar-free scenarios with pretrained models, we find that training only a linear classifier on top of pretrained representations surpasses most continual fine-tuning and prompt-tuning strategies. Overall, our investigation challenges the prevailing assumptions about effective representation learning in the online continual learning.

Xiaoxin He, Yijun Tian, Yifei Sun, Nitesh Chawla, Thomas Laurent, Yann LeCun, Xavier Bresson, Bryan Hooi

Given a graph with textual attributes, we enable users to `chat with their graph': that is, to ask questions about the graph using a conversational interface. In response to a user's questions, our method provides textual replies and highlights the relevant parts of the graph. While existing works integrate large language models (LLMs) and graph neural networks (GNNs) in various ways, they mostly focus on either conventional graph tasks (such as node, edge, and graph classification), or on answering simple graph queries on small or synthetic graphs. In contrast, we develop a flexible question-answering framework targeting real-world textual graphs, applicable to multiple applications including scene graph understanding, common sense reasoning, and knowledge graph reasoning. Toward this goal, we first develop a Graph Question Answering (GraphQA) benchmark with data collected from different tasks. Then, we propose our \textit{G-Retriever} method, introducing the first retrieval-augmented generation (RAG) approach for general textual graphs, which can be fine-tuned to enhance graph understanding via soft prompting. To resist hallucination and to allow for textual graphs that greatly exceed the LLM's context window size, \textit{G-Retriever} performs RAG over a graph by formulating this task as a Prize-Collecting Steiner Tree optimization problem. Empirical evaluations show that our method outperforms baselines on textual graph tasks from multiple domains, scales well with larger graph sizes, and mitigates hallucination.~\footnote{Our codes and datasets are available at: \url{https://github.com/XiaoxinHe/G-Retriever}}

Computer Vision

Ziyu Shan, Yujie Zhang, Yipeng Liu, YILING XU

No-Reference Point Cloud Quality Assessment (NR-PCQA) aims to objectively assess the human perceptual quality of point clouds without relying on pristine-quality point clouds for reference. It is becoming increasingly significant with the rapid advancement of immersive media applications such as virtual reality (VR) and augmented reality (AR). However, current NR-PCQA models attempt to indiscriminately learn point cloud content and distortion representations within a single network, overlooking their distinct contributions to quality information. To address this issue, we propose DisPA, a novel disentangled representation learning framework for NR-PCQA. The framework trains a dual-branch disentanglement network to minimize mutual information (MI) between representations of point cloud content and distortion. Specifically, to fully disentangle representations, the two branches adopt different philosophies: the content-aware encoder is pretrained by a masked auto-encoding strategy, which can allow the encoder to capture semantic information from rendered images of distorted point clouds; the distortion-aware encoder takes a mini-patch map as input, which forces the encoder to focus on low-level distortion patterns. Furthermore, we utilize an MI estimator to estimate the tight upper bound of the actual MI and further minimize it to achieve explicit representation disentanglement. Extensive experimental results demonstrate that DisPA outperforms state-of-the-art methods on multiple PCQA datasets.

Deep Learning · Everything Else

Andi Han, Jiaxiang Li, Wei Huang, Mingyi Hong, Akiko Takeda, Pratik Kumar Jawanpuria, Bamdev Mishra

Large language models (LLMs) have shown impressive capabilities across various tasks. However, training LLMs from scratch requires significant computational power and extensive memory capacity. Recent studies have explored low-rank structures on weights for efficient fine-tuning in terms of parameters and memory, either through low-rank adaptation or factorization. While effective for fine-tuning, low-rank structures are generally less suitable for pretraining because they restrict parameters to a low-dimensional subspace. In this work, we propose to parameterize the weights as a sum of low-rank and sparse matrices for pretraining, which we call SLTrain. The low-rank component is learned via matrix factorization, while for the sparse component, we employ a simple strategy of uniformly selecting the sparsity support at random and learning only the non-zero entries with the fixed support. While being simple, the random fixed-support sparse learning strategy significantly enhances pretraining when combined with low-rank learning. Our results show that SLTrain adds minimal extra parameters and memory costs compared to pretraining with low-rank parameterization, yet achieves substantially better performance, which is comparable to full-rank training. Remarkably, when combined with quantization and per-layer updates, SLTrain can reduce memory requirements by up to 73% when pretraining the LLaMA 7B model.

Optimization · Discrete and Combinatorial Optimization

Yongchun Li, Santanu Dey, Weijun Xie

The classical Canonical Correlation Analysis (CCA) identifies the correlations between two sets of multivariate variables based on their covariance, which has been widely applied in diverse fields such as computer vision, natural language processing, and speech analysis. Despite its popularity, CCA can encounter challenges in explaining correlations between two variable sets within high-dimensional data contexts. Thus, this paper studies Sparse Canonical Correlation Analysis (SCCA) that enhances the interpretability of CCA. We first show that SCCA generalizes three well-known sparse optimization problems, sparse PCA, sparse SVD, and sparse regression, which are all classified as NP-hard problems. This result motivates us to develop strong formulations and efficient algorithms. Our main contributions include (i) the introduction of a combinatorial formulation that captures the essence of SCCA and allows the development of exact and approximation algorithms; (ii) the establishment of the complexity results for two low-rank special cases of SCCA; and (iii) the derivation of an equivalent mixed-integer semidefinite programming model that facilitates a specialized branch-and-cut algorithm with analytical cuts. The effectiveness of our proposed formulations and algorithms is validated through numerical experiments.

Mingbo Hong, Shen Cheng, Haibin Huang, Haoqiang Fan, Shuaicheng Liu

In this paper, we introduce YOLA, a novel framework for object detection in low-light scenarios. Unlike previous works, we propose to tackle this challenging problem from the perspective of feature learning. Specifically, we propose to learn illumination-invariant features through the Lambertian image formation model. We observe that, under the Lambertian assumption, it is feasible to approximate illumination-invariant feature maps by exploiting the interrelationships between neighboring color channels and spatially adjacent pixels. By incorporating additional constraints, these relationships can be characterized in the form of convolutional kernels, which can be trained in a detection-driven manner within a network. Towards this end, we introduce a novel module dedicated to the extraction of illumination-invariant features from low-light images, which can be easily integrated into existing object detection frameworks. Our empirical findings reveal significant improvements in low-light object detection tasks, as well as promising results in both well-lit and over-lit scenarios.

Computer Vision · Segmentation

Zaiquan Yang, Yuhao LIU, Jiaying Lin, Gerhard Hancke, Rynson Lau

This paper explores the weakly-supervised referring image segmentation (WRIS) problem, and focuses on a challenging setup where target localization is learned directly from image-text pairs. We note that the input text description typically already contains detailed information on how to localize the target object, and we also observe that humans often follow a step-by-step comprehension process (\ie, progressively utilizing target-related attributes and relations as cues) to identify the target object. Hence, we propose a novel Progressive Comprehension Network (PCNet) to leverage target-related textual cues from the input description for progressively localizing the target object.Specifically, we first use a Large Language Model (LLM) to decompose the input text description into short phrases. These short phrases are taken as target-related cues and fed into a Conditional Referring Module (CRM) in multiple stages, to allow updating the referring text embedding and enhance the response map for target localization in a multi-stage manner.Based on the CRM, we then propose a Region-aware Shrinking (RaS) loss to constrain the visual localization to be conducted progressively in a coarse-to-fine manner across different stages.Finally, we introduce an Instance-aware Disambiguation (IaD) loss to suppress instance localization ambiguity by differentiating overlapping response maps generated by different referring texts on the same image. Extensive experiments show that our method outperforms SOTA methods on three common benchmarks.

Deep Learning · Theory

Alexander Havrilla, Wenjing Liao

When training deep neural networks, a model's generalization error is often observed to follow a power scaling law dependent both on the model size and the data size. Perhaps the best known example of such scaling laws are for transformer-based large language models (**LLMs**), where networks with billions of parameters are trained on trillions of tokens of text. Yet, despite sustained widespread interest, a rigorous understanding of why transformer scaling laws exist is still missing. To answer this question, we establish novel statistical estimation and mathematical approximation theories for transformers when the input data are concentrated on a low-dimensional manifold. Our theory predicts a power law between the generalization error and both the training data size and the network size for transformers, where the power depends on the intrinsic dimension $d$ of the training data. Notably, the constructed model architecture is shallow, requiring only logarithmic depth in $d$. By leveraging low-dimensional data structures under a manifold hypothesis, we are able to explain transformer scaling laws in a way which respects the data geometry. Moreover, we test our theory with empirical observation by training LLMs on natural language datasets. We find the observed empirical scaling laws closely agree with our theoretical predictions. Taken together, these results rigorously show the intrinsic dimension of data to be a crucial quantity affecting transformer scaling laws in both theory and practice.

Generative Models · Diffusion models

Haotian Ye, Haowei Lin, Jiaqi Han, Minkai Xu, Sheng Liu, Yitao Liang, Jianzhu Ma, James Zou, Stefano Ermon

Given an unconditional diffusion model and a predictor for a target property of interest (e.g., a classifier), the goal of training-free guidance is to generate samples with desirable target properties without additional training. Existing methods, though effective in various individual applications, often lack theoretical grounding and rigorous testing on extensive benchmarks. As a result, they could even fail on simple tasks, and applying them to a new problem becomes unavoidably difficult. This paper introduces a novel algorithmic framework encompassing existing methods as special cases, unifying the study of training-free guidance into the analysis of an algorithm-agnostic design space. Via theoretical and empirical investigation, we propose an efficient and effective hyper-parameter searching strategy that can be readily applied to any downstream task. We systematically benchmark across 7 diffusion models on 16 tasks with 40 targets, and improve performance by 8.5% on average. Our framework and benchmark offer a solid foundation for conditional generation in a training-free manner.

Language · Speech

Dongchao Yang, Haohan Guo, Yuanyuan Wang, Rongjie Huang, Xiang Li, Xu Tan, Xixin Wu, Helen Meng

Large Language models (LLMs) have demonstrated supreme capabilities in textual understanding and generation, but cannot be directly applied to cross-modal tasks without fine-tuning. This paper proposes a cross-modal in-context learning approach, empowering the frozen LLMs to achieve multiple audio tasks in a few-shot style without any parameter update. Specifically, we propose a novel LLM-driven audio codec model, LLM-Codec, which transfers the audio modality into textual space by representing audio tokens with words or sub-words from the LLM vocabulary, while maintaining high audio reconstruction quality.The key idea is to reduce the modality heterogeneity between text and audio by compressing the audio modality into the well-trained textual space of LLMs. Thus, the audio representation can be viewed as a new \textit{foreign language}, and LLMs can learn the new \textit{foreign language} with several demonstrations. In experiments, we investigate the performance of the proposed approach across multiple audio understanding and generation tasks, \textit{e.g.} speech emotion classification, audio classification, text-to-speech generation, speech enhancement, etc. Experimental results show that LLMs equipped with the LLM-Codec, named as UniAudio 1.5, prompted by only a few examples, can perform effectively in simple scenarios, validating our cross-modal in-context learning approach.To facilitate research on few-shot audio task learning and multi-modal LLMs, we have open-sourced the LLM-Codec model.

Haoxuan Li, Yue Liu, Zhi Geng, Kun Zhang

Developing fair automated machine learning algorithms is critical in making safe and trustworthy decisions. Many causality-based fairness notions have been proposed to address the above issues by quantifying the causal connections between sensitive attributes and decisions, and when the true causal graph is fully known, certain algorithms that achieve interventional fairness have been proposed. However, when the true causal graph is unknown, it is still challenging to effectively and efficiently exploit partially directed acyclic graphs (PDAGs) to achieve interventional fairness. To exploit the PDAGs for achieving interventional fairness, previous methods have been built on variable selection or causal effect identification, but limited to reduced prediction accuracy or strong assumptions. In this paper, we propose a general min-max optimization framework that can achieve interventional fairness with promising prediction accuracy and can be extended to maximally oriented PDAGs (MPDAGs) with added background knowledge. Specifically, we first estimate all possible treatment effects of sensitive attributes on a given prediction model from all possible adjustment sets of sensitive attributes via an efficient local approach. Next, we propose to alternatively update the prediction model and possible estimated causal effects, where the prediction model is trained via a min-max loss to control the worst-case fairness violations. Extensive experiments on synthetic and real-world datasets verify the superiority of our methods. To benefit the research community, we have released our project at https://github.com/haoxuanli-pku/NeurIPS24-Interventional-Fairness-with-PDAGs.

Generative Models

Minghan Li, Xilun Chen, Ari Holtzman, Beidi Chen, Jimmy Lin, Scott Yih, Victoria Lin

Large language models (LLMs) often hallucinate and lack the ability to provide attribution for their generations. Semi-parametric LMs, such as kNN-LM, approach these limitations by refining the output of an LM for a given prompt using its nearest neighbor matches in a non-parametric data store. However, these models often exhibit slow inference speeds and produce non-fluent texts. In this paper, we introduce Nearest Neighbor Speculative Decoding (NEST), a novel semi-parametric language modeling approach that is capable of incorporating real-world text spans of arbitrary length into the LM generations and providing attribution to their sources. NEST performs token-level retrieval at each inference step to compute a semi-parametric mixture distribution and identify promising span continuations in a corpus. It then uses an approximate speculative decoding procedure that accepts a prefix of the retrieved span or generates a new token. NEST significantly enhances the generation quality and attribution rate of the base LM across a variety of knowledge-intensive tasks, surpassing the conventional kNN-LM method and performing competitively with in-context retrieval augmentation. In addition, NEST substantially improves the generation speed, achieving a 1.8x speedup in inference time when applied to Llama-2-Chat 70B. Code will be released at https://github.com/facebookresearch/NEST/tree/main.

Applications

Che Liu, Cheng Ouyang, Sibo Cheng, Anand Shah, Wenjia Bai, Rossella Arcucci

Medical imaging tasks require an understanding of subtle and localized visual features due to the inherently detailed and area-specific nature of pathological patterns, which are crucial for clinical diagnosis. Although recent advances in medical vision-language pre-training (VLP) enable models to learn clinically relevant visual features by leveraging both medical images and their associated radiology reports, current medical VLP methods primarily focus on aligning images with entire reports. This focus hinders the learning of dense (pixel-level) visual features and is suboptimal for dense prediction tasks (e.g., medical image segmentation).To address this challenge, we propose a novel medical VLP framework, named **Global to Dense level representation learning (G2D)**, which aims to learn global and dense visual features simultaneously using only image-text pairs without extra annotations. In particular, G2D designs a **Pseudo Segmentation (PS)** task, which enables the model to learn dense visual features during VLP. Notably, generating PS masks can be performed on the fly during VLP, which does not incur extra trainable parameters. With this simple yet effective idea, G2D achieves superior performance across 5 medical imaging tasks and 25 diseases. Particularly, in the segmentation task which requires dense visual features, **G2D surpasses existing models even with just 1% of the training data for finetuning, compared to 100% used by other models**. The code can be found in https://github.com/cheliu-computation/G2D-NeurIPS24/tree/main.

Multimodal Models

Bing Cao, Yinan Xia, Yi Ding, Changqing Zhang, Qinghua Hu

The inherent challenge of image fusion lies in capturing the correlation of multi-source images and comprehensively integrating effective information from different sources. Most existing techniques fail to perform dynamic image fusion while notably lacking theoretical guarantees, leading to potential deployment risks in this field. Is it possible to conduct dynamic image fusion with a clear theoretical justification? In this paper, we give our solution from a generalization perspective. We proceed to reveal the generalized form of image fusion and derive a new test-time dynamic image fusion paradigm. It provably reduces the upper bound of generalization error. Specifically, we decompose the fused image into multiple components corresponding to its source data. The decomposed components represent the effective information from the source data, thus the gap between them reflects the \textit{Relative Dominability} (RD) of the uni-source data in constructing the fusion image. Theoretically, we prove that the key to reducing generalization error hinges on the negative correlation between the RD-based fusion weight and the uni-source reconstruction loss. Intuitively, RD dynamically highlights the dominant regions of each source and can be naturally converted to the corresponding fusion weight, achieving robust results. Extensive experiments and discussions with in-depth analysis on multiple benchmarks confirm our findings and superiority. Our code is available at https://github.com/Yinan-Xia/TTD.

Language · Factuality

Yuzhe Gu, Ziwei Ji, Wenwei Zhang, Chengqi Lyu, Dahua Lin, Kai Chen

Large language models (LLMs) exhibit hallucinations in long-form question-answering tasks across various domains and wide applications. Current hallucination detection and mitigation datasets are limited in domain and size, which struggle to scale due to prohibitive labor costs and insufficient reliability of existing hallucination annotators. To facilitate the scalable oversight of LLM hallucinations, this paper introduces an iterative self-training framework that simultaneously and progressively scales up the annotation dataset and improves the accuracy of the annotator. Based on the Expectation Maximization algorithm, in each iteration, the framework first applies an automatic hallucination annotation pipeline for a scaled dataset and then trains a more accurate annotator on the dataset. This new annotator is adopted in the annotation pipeline for the next iteration. Extensive experimental results demonstrate that the finally obtained hallucination annotator with only 7B parameters surpasses GPT-4 and obtains new state-of-the-art hallucination detection results on HaluEval and HalluQA by zero-shot inference. Such an annotator can not only evaluate the hallucination levels of various LLMs on the large-scale dataset but also help to mitigate the hallucination of LLMs generations, with the Natural Language Inference metric increasing from 25% to 37% on HaluEval.

Computer Vision

Xingyu Zhu, Beier Zhu, Yi Tan, Shuo Wang, Yanbin Hao, Hanwang Zhang

Vision-language models, such as CLIP, have shown impressive generalization capacities when using appropriate text descriptions. While optimizing prompts on downstream labeled data has proven effective in improving performance, these methods entail labor costs for annotations and are limited by their quality. Additionally, since CLIP is pre-trained on highly imbalanced Web-scale data, it suffers from inherent label bias that leads to suboptimal performance. To tackle the above challenges, we propose a label-**F**ree p**ro**mpt distribution **l**earning and b**i**as **c**orrection framework, dubbed as **Frolic**, which boosts zero-shot performance without the need for labeled data. Specifically, our Frolic learns distributions over prompt prototypes to capture diverse visual representations and adaptively fuses these with the original CLIP through confidence matching.This fused model is further enhanced by correcting label bias via a label-free logit adjustment. Notably, our method is not only training-free but also circumvents the necessity for hyper-parameter tuning. Extensive experimental results across 16 datasets demonstrate the efficacy of our approach, particularly outperforming the state-of-the-art by an average of $2.6\%$ on 10 datasets with CLIP ViT-B/16 and achieving an average margin of $1.5\%$ on ImageNet and its five distribution shifts with CLIP ViT-B/16. Codes are available in [https://github.com/zhuhsingyuu/Frolic](https://github.com/zhuhsingyuu/Frolic).

Hanchen Xia, Weidong Liu, Xiaojun Mao

The cost of ranking becomes significant in the new stage of deep learning. We propose ST$_k$, a fully differentiable module with a single trainable parameter, designed to solve the Top-k problem without requiring additional time or GPU memory. Due to its fully differentiable nature, ST$_k$ can be embedded end-to-end into neural networks and optimize the Top-k problems within a unified computational graph. We apply ST$_k$ to the Average Top-k Loss (AT$_k$), which inherently faces a Top-k problem. The proposed ST$_k$ Loss outperforms AT$_k$ Loss and achieves the best average performance on multiple benchmarks, with the lowest standard deviation. With the assistance of ST$_k$ Loss, we surpass the state-of-the-art (SOTA) on both CIFAR-100-LT and Places-LT leaderboards.

Aristeidis Panos

The standard paradigm of modeling marked point processes is by parameterizing the intensity function using an attention-based (Transformer-style) architecture. Despite the flexibility of these methods, their inference is based on the computationally intensive thinning algorithm. In this work, we propose a framework where the advantages of the attention-based architecture are maintained and the limitation of the thinning algorithm is circumvented. The framework depends on modeling the conditional distribution of inter-event times with a mixture of log-normals satisfying a Markov property and the conditional probability mass function for the marks with a Transformer-based architecture. The proposed method attains state-of-the-art performance in predicting the next event of a sequence given its history. The experiments also reveal the efficacy of the methods that do not rely on the thinning algorithm during inference over the ones they do. Finally, we test our method on the challenging long-horizon prediction task and find that it outperforms a baseline developed specifically for tackling this task; importantly, inference requires just a fraction of time compared to the thinning-based baseline.

Yangjun Ruan, Chris Maddison, Tatsunori Hashimoto

Understanding how language model performance varies with scale is critical to benchmark and algorithm development. Scaling laws are one approach to building this understanding, but the requirement of training models across many different scales has limited their use. We propose an alternative, observational approach that bypasses model training and instead builds scaling laws from ~100 publically available models. Building a single scaling law from multiple model families is challenging due to large variations in their training compute efficiencies and capabilities. However, we show that these variations are consistent with a simple, generalized scaling law where language model performance is a function of a low-dimensional capability space, and model families only vary in their efficiency in converting training compute to capabilities. Using this approach, we show the surprising predictability of complex scaling phenomena: we show that several emergent phenomena follow a smooth, sigmoidal behavior and are predictable from small models; we show that the agent performance of models such as GPT-4 can be precisely predicted from simpler non-agentic benchmarks; and we show how to predict the impact of post-training interventions like Chain-of-Thought and Self-Consistency as language model capabilities continue to improve.

Privacy

Saptarshi Roy, Zehua Wang, Ambuj Tewari

We consider the problem of model selection in a high-dimensional sparse linear regression model under privacy constraints. We propose a differentially private (DP) best subset selection method with strong statistical utility properties by adopting the well-known exponential mechanism for selecting the best model. To achieve computational expediency, we propose an efficient Metropolis-Hastings algorithm and under certain regularity conditions, we establish that it enjoys polynomial mixing time to its stationary distribution. As a result, we also establish both approximate differential privacy and statistical utility for the estimates of the mixed Metropolis-Hastings chain. Finally, we perform some illustrative experiments on simulated data showing that our algorithm can quickly identify active features under reasonable privacy budget constraints.