Towards Privacy-Guaranteed Label Unlearning in Vertical Federated Learning: Few-Shot Forgetting without Disclosure

The field of artificial intelligence (AI) has witnessed significant advancements in recent years, transforming the way we approach complex problems and interact with technology. However, despite these breakthroughs, several challenges persist, hindering the widespread adoption of AI in various industries. Five recent studies, published on arXiv, tackle some of these challenges head-on, offering innovative solutions to pressing issues in machine learning.

One of the primary concerns in AI research is data privacy, particularly in the context of federated learning. A study titled "Towards Privacy-Guaranteed Label Unlearning in Vertical Federated Learning" proposes a novel method for label unlearning in vertical federated learning (VFL), a setting that has received relatively little attention compared to its horizontal counterpart. The authors introduce a representation-level manifold mixup mechanism to generate synthetic embeddings for both unlearned and retained samples, effectively removing sensitive label information from the model while maintaining its performance on retained data.

Another study, "Beyond Attribution: Unified Concept-Level Explanations," addresses the need for more comprehensive explanation techniques in AI models. The authors propose a general framework called UnCLE, which elevates existing local model-agnostic techniques to provide concept-based explanations. UnCLE can uniformly extend existing local model-agnostic methods to provide unified concept-based explanations with large pre-trained model perturbation, making it a valuable tool for understanding complex AI models.

In the realm of image synthesis, few-step diffusion models have shown promise, but they struggle to align with specific downstream objectives. A study titled "Aligning Few-Step Diffusion Models with Dense Reward Difference Learning" proposes a novel reinforcement learning (RL) framework called Stepwise Diffusion Policy Optimization (SDPO). SDPO introduces a dual-state trajectory sampling mechanism and a latent similarity-based dense reward prediction strategy to enable low-variance, mixed-step optimization, making it more efficient and effective.

Neuro-symbolic AI, which combines neural networks with symbolic reasoning, has been gaining traction in recent years. A study titled "Neuro-Symbolic AI for Analytical Solutions of Differential Equations" presents a framework called SIGS, which automates the process of discovering analytical solutions to differential equations. SIGS uses a formal grammar to generate syntactically valid building blocks, embeds these expressions into a continuous space, and then searches this space to assemble, score, and refine candidate closed-form solutions.

Lastly, a study titled "Mixing It Up: Exploring Mixer Networks for Irregular Multivariate Time Series Forecasting" proposes a novel architecture called IMTS-Mixer, which adapts the principles of Mixer models to the irregular multivariate time series (IMTS) setting. IMTS-Mixer introduces two key components: a channel-wise encoder that transforms irregular observations into fixed-size vectors and a continuous time decoder that supports forecasting at arbitrary time points.

These studies demonstrate the breadth and depth of AI research, tackling challenges in data privacy, explanation techniques, image synthesis, neuro-symbolic AI, and time series forecasting. As AI continues to transform industries and revolutionize the way we interact with technology, it is essential to address these challenges and push the boundaries of what's possible in machine learning.

Sources:

• "Towards Privacy-Guaranteed Label Unlearning in Vertical Federated Learning" (arXiv:2410.10922v3)
• "Beyond Attribution: Unified Concept-Level Explanations" (arXiv:2410.12439v2)
• "Aligning Few-Step Diffusion Models with Dense Reward Difference Learning" (arXiv:2411.11727v2)
• "Neuro-Symbolic AI for Analytical Solutions of Differential Equations" (arXiv:2502.01476v3)
• "Mixing It Up: Exploring Mixer Networks for Irregular Multivariate Time Series Forecasting" (arXiv:2502.11816v3)