GRU: Mitigating the Trade-off between Unlearning and Retention for Large Language Models

Large language model (LLM) unlearning has demonstrated its essential role in removing privacy and copyright-related responses, crucial for their legal and safe applications. However, the pursuit of complete unlearning often comes with substantial costs due to its compromises in their general functionality, leading to a notorious trade-off between unlearning and retention. In examining the update process for unlearning dynamically, we find gradients hold essential information for revealing this trade-off. In particular, we look at the varying relationship between retention performance and directional disparities between gradients during unlearning. It motivates the sculpting of an update mechanism derived from gradients from two sources, i.e., harmful for retention and useful for unlearning. Accordingly, we propose Gradient Rectified Unlearning (GRU), an enhanced unlearning framework controlling the updating gradients in a geometry-focused and optimization-driven manner such that their side impacts on other, unrelated responses can be minimized. Specifically, GRU derives a closed-form solution to project the unlearning gradient onto the orthogonal space of that gradient harmful for retention, ensuring minimal deviation from its original direction under the condition that overall performance is retained. Comprehensive experiments are conducted to demonstrate that GRU, as a general framework, is straightforward to implement and efficiently enhances a range of baseline methods through its adaptable and compatible characteristics. Additionally, experimental results show its broad effectiveness across a diverse set of benchmarks for LLM unlearning.

Rethinking LLM Unlearning Objectives: A Gradient Perspective and Go Beyond

Large language models (LLMs) should undergo rigorous audits to identify potential risks, such as copyright and privacy infringements. Once these risks emerge, timely updates are crucial to remove undesirable responses, ensuring legal and safe model usage. It has spurred recent research into LLM unlearning, focusing on erasing targeted undesirable knowledge without compromising the integrity of other, non-targeted responses. Existing studies have introduced various unlearning objectives to pursue LLM unlearning without necessitating complete retraining. However, each of these objectives has unique properties, and no unified framework is currently available to comprehend them thoroughly. To fill the gap, we propose a toolkit of the gradient effect (G-effect), quantifying the impacts of unlearning objectives on model performance from a gradient perspective. A notable advantage is its broad ability to detail the unlearning impacts from various aspects across instances, updating steps, and LLM layers. Accordingly, the G-effect offers new insights into identifying drawbacks of existing unlearning objectives, further motivating us to explore a series of new solutions for their mitigation and improvements. Finally, we outline promising directions that merit further studies, aiming at contributing to the community to advance this important field.

GenIAS: Generator for Instantiating Anomalies in time Series

A recent and promising approach for building time series anomaly detection (TSAD) models is to inject synthetic samples of anomalies within real data sets. The existing injection mechanisms have significant limitations - most of them rely on ad hoc, hand-crafted strategies which fail to capture the natural diversity of anomalous patterns, or are restricted to univariate time series settings. To address these challenges, we design a generative model for TSAD using a variational autoencoder, which is referred to as a Generator for Instantiating Anomalies in Time Series (GenIAS). GenIAS is designed to produce diverse and realistic synthetic anomalies for TSAD tasks. By employing a novel learned perturbation mechanism in the latent space and injecting the perturbed patterns in different segments of time series, GenIAS can generate anomalies with greater diversity and varying scales. Further, guided by a new triplet loss function, which uses a min-max margin and a new variance-scaling approach to further enforce the learning of compact normal patterns, GenIAS ensures that anomalies are distinct from normal samples while remaining realistic. The approach is effective for both univariate and multivariate time series. We demonstrate the diversity and realism of the generated anomalies. Our extensive experiments demonstrate that GenIAS - when integrated into a TSAD task - consistently outperforms seventeen traditional and deep anomaly detection models, thereby highlighting the potential of generative models for time series anomaly generation.

Unlearning with Control: Assessing Real-world Utility for Large Language Model Unlearning

The compelling goal of eradicating undesirable data behaviors, while preserving usual model functioning, underscores the significance of machine unlearning within the domain of large language models (LLMs). Recent research has begun to approach LLM unlearning via gradient ascent (GA) -- increasing the prediction risk for those training strings targeted to be unlearned, thereby erasing their parameterized responses. Despite their simplicity and efficiency, we suggest that GA-based methods face the propensity towards excessive unlearning, resulting in various undesirable model behaviors, such as catastrophic forgetting, that diminish their practical utility. In this paper, we suggest a set of metrics that can capture multiple facets of real-world utility and propose several controlling methods that can regulate the extent of excessive unlearning. Accordingly, we suggest a general framework to better reflect the practical efficacy of various unlearning methods -- we begin by controlling the unlearning procedures/unlearned models such that no excessive unlearning occurs and follow by the evaluation for unlearning efficacy. Our experimental analysis on established benchmarks revealed that GA-based methods are far from perfect in practice, as strong unlearning is at the high cost of hindering the model utility. We conclude that there is still a long way towards practical and effective LLM unlearning, and more efforts are required in this field.

Do CLIPs Always Generalize Better than ImageNet Models?

Large vision language models, such as CLIPs, have revolutionized modern machine learning. CLIPs have demonstrated great generalizability under distribution shifts, supported by an increasing body of literature. However, the evaluation datasets for CLIPs are variations primarily designed for ImageNet benchmarks, which may not fully reflect the extent to which CLIPs, e.g., pre-trained on LAION, robust to spurious correlations. To bridge the gap, we collect a real-world dataset called CounterAnimal that contains realistic spurious features found in animal photos. CounterAnimal consists of a) the common group: comprising animals on common backgrounds, and b) the counter group: including animals on unusual backgrounds. The performance drops from the common to counter groups quantify the reliance of models on spurious features (i.e., backgrounds) to predict the animals. We find that CLIPs trained on either LAION or the OpenAI data exhibit notable performance drops on the counter group. Surprisingly, we observe that single-modal models trained on ImageNet are more robust than CLIPs. We provide both theoretical and empirical explanations for why CLIPs still learn spurious features. Our findings suggest that distribution shifts remain an open problem for CLIPs, and one needs to be cautious about test setups when evaluating foundation models pre-trained on a significantly different scale and distribution.

Are All Unseen Data Out-of-Distribution?

Distributions of unseen data have been all treated as out-of-distribution (OOD), making their generalization a significant challenge. Much evidence suggests that the size increase of training data can monotonically decrease generalization errors in test data. However, this is not true from other observations and analysis. In particular, when the training data have multiple source domains and the test data contain distribution drifts, then not all generalization errors on the test data decrease monotonically with the increasing size of training data. Such a non-decreasing phenomenon is formally investigated under a linear setting with empirical verification across varying visual benchmarks. Motivated by these results, we redefine the OOD data as a type of data outside the convex hull of the training domains and prove a new generalization bound based on this new definition. It implies that the effectiveness of a well-trained model can be guaranteed for the unseen data that is within the convex hull of the training domains. But, for some data beyond the convex hull, a non-decreasing error trend can happen. Therefore, we investigate the performance of popular strategies such as data augmentation and pre-training to overcome this issue. Moreover, we propose a novel reinforcement learning selection algorithm in the source domains only that can deliver superior performance over the baseline methods.

Artificial intelligence optical hardware empowers high-resolution hyperspectral video understanding at 1.2 Tb/s

Foundation models, exemplified by GPT technology, are discovering new horizons in artificial intelligence by executing tasks beyond their designers' expectations. While the present generation provides fundamental advances in understanding language and images, the next frontier is video comprehension. Progress in this area must overcome the 1 Tb/s data rate demanded to grasp real-time multidimensional video information. This speed limit lies well beyond the capabilities of the existing generation of hardware, imposing a roadblock to further advances. This work introduces a hardware-accelerated integrated optoelectronic platform for multidimensional video understanding in real-time. The technology platform combines artificial intelligence hardware, processing information optically, with state-of-the-art machine vision networks, resulting in a data processing speed of 1.2 Tb/s with hundreds of frequency bands and megapixel spatial resolution at video rates. Such performance, validated in the AI tasks of video semantic segmentation and object understanding in indoor and aerial applications, surpasses the speed of the closest technologies with similar spectral resolution by three to four orders of magnitude. This platform opens up new avenues for research in real-time AI video understanding of multidimensional visual information, helping the empowerment of future human-machine interactions and cognitive processing developments.

Open-Set Graph Anomaly Detection via Normal Structure Regularisation

This paper considers an under-explored Graph Anomaly Detection (GAD) task, namely open-set GAD, which aims to detect anomalous nodes using a small number of labelled training normal and anomaly nodes (known as seen anomalies) that cannot illustrate all possible inference-time abnormalities. The task has attracted growing attention due to the availability of anomaly prior knowledge from the label information that can help to substantially reduce detection errors. However, current methods tend to over-emphasise fitting the seen anomalies, leading to a weak generalisation ability to detect unseen anomalies, i.e., those that are not illustrated by the labelled anomaly nodes. Further, they were introduced to handle Euclidean data, failing to effectively capture important non-Euclidean features for GAD. In this work, we propose a novel open-set GAD approach, namely normal structure regularisation (NSReg), to leverage the rich normal graph structure embedded in the labelled nodes to tackle the aforementioned two issues. In particular, NSReg trains an anomaly-discriminative supervised graph anomaly detector, with a plug-and-play regularisation term to enforce compact, semantically-rich representations of normal nodes. To this end, the regularisation is designed to differentiate various types of normal nodes, including labelled normal nodes that are connected in their local neighbourhood, and those that are not connected. By doing so, it helps incorporate strong normality into the supervised anomaly detector learning, mitigating their overfitting to the seen anomalies. Extensive empirical results on real-world datasets demonstrate the superiority of our proposed NSReg for open-set GAD.

Out-of-distribution Detection Learning with Unreliable Out-of-distribution Sources

Out-of-distribution (OOD) detection discerns OOD data where the predictor cannot make valid predictions as in-distribution (ID) data, thereby increasing the reliability of open-world classification. However, it is typically hard to collect real out-of-distribution (OOD) data for training a predictor capable of discerning ID and OOD patterns. This obstacle gives rise to data generation-based learning methods, synthesizing OOD data via data generators for predictor training without requiring any real OOD data. Related methods typically pre-train a generator on ID data and adopt various selection procedures to find those data likely to be the OOD cases. However, generated data may still coincide with ID semantics, i.e., mistaken OOD generation remains, confusing the predictor between ID and OOD data. To this end, we suggest that generated data (with mistaken OOD generation) can be used to devise an auxiliary OOD detection task to facilitate real OOD detection. Specifically, we can ensure that learning from such an auxiliary task is beneficial if the ID and the OOD parts have disjoint supports, with the help of a well-designed training procedure for the predictor. Accordingly, we propose a powerful data generation-based learning method named Auxiliary Task-based OOD Learning (ATOL) that can relieve the mistaken OOD generation. We conduct extensive experiments under various OOD detection setups, demonstrating the effectiveness of our method against its advanced counterparts.

Learning to Augment Distributions for Out-of-Distribution Detection

Open-world classification systems should discern out-of-distribution (OOD) data whose labels deviate from those of in-distribution (ID) cases, motivating recent studies in OOD detection. Advanced works, despite their promising progress, may still fail in the open world, owing to the lack of knowledge about unseen OOD data in advance. Although one can access auxiliary OOD data (distinct from unseen ones) for model training, it remains to analyze how such auxiliary data will work in the open world. To this end, we delve into such a problem from a learning theory perspective, finding that the distribution discrepancy between the auxiliary and the unseen real OOD data is the key to affecting the open-world detection performance. Accordingly, we propose Distributional-Augmented OOD Learning (DAL), alleviating the OOD distribution discrepancy by crafting an OOD distribution set that contains all distributions in a Wasserstein ball centered on the auxiliary OOD distribution. We justify that the predictor trained over the worst OOD data in the ball can shrink the OOD distribution discrepancy, thus improving the open-world detection performance given only the auxiliary OOD data. We conduct extensive evaluations across representative OOD detection setups, demonstrating the superiority of our DAL over its advanced counterparts.