In recent years, low-rank adaptation (LoRA) has emerged as a significant paradigm that freezes pre-trained weights and introduces small, learnable adapters instead of fine-tuning the full set of parameters.In this work, we uncover several key insights regarding the singular components of network parameters based on Singular Value Decomposition (SVD).Firstly, the principal singular components with large singular values in pre-trained network parameters can be effectively reused during finetuning, whereas the minor components with smaller singular values are more task-specific and require substantial adaptation.Secondly, we first establish the theoretical connection that the uncontrolled growth of singular values in LoRA adapters leads to the forgetting of pretrained knowledge -a well-known issue referred to as catastrophic forgetting.Building on these observations, we propose SCLoRA, which injects parameterized singular components with spectral clipping into the pre-trained model in a way that is aware of the spectral distribution of the pre-trained model.SCLoRA effectively adapts to new tasks by focusing updates on components that require adaptation, while simultaneously alleviating catastrophic forgetting.We conduct extensive experiments and demonstrate that SCLoRA not only improves downstream performance but also effectively retains pre-trained knowledge.

Transformers have demonstrated remarkable performance across diverse domains. The key component of Transformers is self-attention, which learns the relationship between any two tokens in the input sequence. Recent studies have revealed that the self-attention can be understood as a normalized adjacency matrix of a graph. Notably, from the perspective of graph signal processing (GSP), the self-attention can be equivalently defined as a simple graph filter, applying GSP using the value vector as the signal. However, the self-attention is a graph filter defined with only the first order of the polynomial matrix, and acts as a low-pass filter preventing the effective leverage of various frequency information. Consequently, existing self-attention mechanisms are designed in a rather simplified manner. Therefore, we propose a novel method, called Attentive Graph Filter (AGF), interpreting the self-attention as learning the graph filter in the singular value domain from the perspective of graph signal processing for directed graphs with the linear complexity w.r.t. the input length. In our experiments, we demonstrate that AGF achieves state-of-the-art performance on various tasks, including Long Range Arena benchmark and time series classification. Code is available at https://github.com/hyowonwi/agf.

Sequential recommendation (SR) models based on Transformers have achieved remarkable successes. The self-attention mechanism of Transformers for computer vision and natural language processing suffers from the oversmoothing problem, i.e., hidden representations becoming similar to tokens. In the SR domain, we, for the first time, show that the same problem occurs. We present pioneering investigations that reveal the low-pass filtering nature of self-attention in the SR, which causes oversmoothing. To this end, we propose a novel method called Beyond Self-Attention for Sequential Recommendation (BSARec), which leverages the Fourier transform to i) inject an inductive bias by considering fine-grained sequential patterns and ii) integrate low and high-frequency information to mitigate oversmoothing. Our discovery shows significant advancements in the SR domain and is expected to bridge the gap for existing Transformer-based SR models. We test our proposed approach through extensive experiments on 6 benchmark datasets. The experimental results demonstrate that our model outperforms 7 baseline methods in terms of recommendation performance. Our code is available at https://github.com/yehjin-shin/BSARec.

Recent research in the field of graph neural network (GNN) has identified a critical issue known as "over-squashing," resulting from the bottleneck phenomenon in graph structures, which impedes the propagation of long-range information. Prior works have proposed a variety of graph rewiring concepts that aim at optimizing the spatial or spectral properties of graphs to promote the signal propagation. However, such approaches inevitably deteriorate the original graph topology, which may lead to a distortion of information flow. To address this, we introduce an ex**pand**ed width-**a**ware (**PANDA**) message passing, a new message passing paradigm where nodes with high centrality, a potential source of over-squashing, are selectively expanded in width to encapsulate the growing influx of signals from distant nodes. Experimental results show that our method outperforms existing rewiring methods, suggesting that selectively expanding the hidden state of nodes can be a compelling alternative to graph rewiring for addressing the over-squashing.

Structured data, which constitutes a significant portion of existing data types, has been a long-standing research topic in the field of machine learning. Various representation learning methods for tabular data have been proposed, ranging from encoder-decoder structures to Transformers. Among these, Transformer-based methods have achieved state-of-the-art performance not only in tabular data but also in various other fields, including computer vision and natural language processing. However, recent studies have revealed that self-attention, a key component of Transformers, can lead to an oversmoothing issue. We show that Transformers for tabular data also face this problem. To tackle the problem, we suggest a novel self-attention layer for tabular data, leveraging matrix polynomials. This proposed layer serves as a replacement for the original self-attention layer, contributing to the improvement of model scalability. In our experiments with three representative table learning models equipped with our proposed layer, we illustrate that the layer effectively mitigates the oversmoothing problem and enhances the representation performance of the existing methods, outperforming the state-of-the-art table representation methods.

Transformers, renowned for their self-attention mechanism, have achieved state-of-the-art performance across various tasks in natural language processing, computer vision, time-series modeling, etc. However, one of the challenges with deep Transformer models is the oversmoothing problem, where representations across layers converge to indistinguishable values, leading to significant performance degradation. We interpret the original self-attention as a simple graph filter and redesign it from a graph signal processing (GSP) perspective. We propose a graph-filter-based self-attention (GFSA) to learn a general yet effective one, whose complexity, however, is slightly larger than that of the original self-attention mechanism. We demonstrate that GFSA improves the performance of Transformers in various fields, including computer vision, natural language processing, graph-level tasks, speech recognition, and code classification.

Time series imputation is one of the most fundamental tasks for time series. Real-world time series datasets are frequently incomplete (or irregular with missing observations), in which case imputation is strongly required. Many different time series imputation methods have been proposed. Recent self-attention-based methods show the state-of-the-art imputation performance. However, it has been overlooked for a long time to design an imputation method based on continuous-time recurrent neural networks (RNNs), i.e., neural controlled differential equations (NCDEs). To this end, we redesign time series (variational) autoencoders based on NCDEs. Our method, called continuous-time autoencoder (CTA), encodes an input time series sample into a continuous hidden path (rather than a hidden vector) and decodes it to reconstruct and impute the input. In our experiments with 4 datasets and 19 baselines, our method shows the best imputation performance in almost all cases.

Long-term time series forecasting (LTSF) is a challenging task that has been investigated in various domains such as finance investment, health care, traffic, and weather forecasting. In recent years, Linear-based LTSF models showed better performance, pointing out the problem of Transformer-based approaches causing temporal information loss. However, Linear-based approach has also limitations that the model is too simple to comprehensively exploit the characteristics of the dataset. To solve these limitations, we propose LTSF-DNODE, which applies a model based on linear ordinary differential equations (ODEs) and a time series decomposition method according to data statistical characteristics. We show that LTSF-DNODE outperforms the baselines on various real-world datasets. In addition, for each dataset, we explore the impacts of regularization in the neural ordinary differential equation (NODE) framework.

Recommender systems are a long-standing research problem in data mining and machine learning. They are incremental in nature, as new user-item interaction logs arrive. In real-world applications, we need to periodically train a collaborative filtering algorithm to extract user/item embedding vectors and therefore, a time-series of embedding vectors can be naturally defined. We present a time-series forecasting-based upgrade kit (TimeKit), which works in the following way: it i) first decides a base collaborative filtering algorithm, ii) extracts user/item embedding vectors with the base algorithm from user-item interaction logs incrementally, e.g., every month, iii) trains our time-series forecasting model with the extracted time-series of embedding vectors, and then iv) forecasts the future embedding vectors and recommend with their dot-product scores owing to a recent breakthrough in processing complicated time-series data, i.e., neural controlled differential equations (NCDEs). Our experiments with four real-world benchmark datasets show that the proposed time-series forecasting-based upgrade kit can significantly enhance existing popular collaborative filtering algorithms.