29 accepted papers
Large Language Models (LLMs) powerful generative capabilities also pose significant risks, underscoring the need for effective detoxification methods to ensure safer deployment. Due to the polysemantic nature of LLM neurons, recent neuron intervention methods inevitably entangle unrelated concepts,
The impressive performance of large language models (LLMs) also brings inherent toxicity risks, prompting the need for effective detoxification to support responsible deployment. Prevailing methods generally follow an inflexible model-specific fashion, addressing only individual models or model fami
There has been significant recent interest on understanding the capacity of Transformers for in-context learning (ICL), yet most theory focuses on supervised settings with explicitly labeled pairs. In practice, Transformers often perform well even when labels are sparse or absent, suggesting crucial…
Fine-tuning Large Language Models (LLMs) enables data holders to construct proprietary, task-specific models by leveraging external high-performance computing infrastructure. However, existing paradigms typically address data privacy and model intellectual property (IP) in isolation, failing to simu…
Steering Vector (SV) is a powerful technique for controlling Large Language Models (LLMs) by manipulating their activations without altering model weights. However, when constructed from sensitive data, SV poses significant privacy risks, as it may leak private information. Existing differential pri
Fixed-position grasping is an efficient strategy in robotic assembly. Adaptive grippers employing differential mechanisms (DMs) can passively compensate for misalignment between gripper and object. However, conventional DMs, such as gear-based differential mechanisms (GDMs), are limited by rotationa
While Vision Language Models (VLMs) have demonstrated remarkable capabilities in general visual understanding, their application in the chemical domain has been limited, with previous works predominantly focusing on text and thus overlooking critical visual information, such as molecular structures.
In organic reaction prediction, many recent approaches ranging from traditional task-specific models to Large Language Models (LLMs), have demonstrated notable success. However, these methods are inherently data-driven, exhibit constrained interpretability, and have hit fundamental performance bottl…
Event cameras provide microsecond latency and high dynamic range, making them ideal for 3D perception tasks in traffic scenes with challenging lighting conditions. Yet existing methods often struggle to generalize to out-of-domain environments due to the limited availability of diverse training data
Prior-data fitted networks (PFNs) have recently emerged as a powerful approach for Bayesian prediction tasks, approximating the posterior predictive distribution (PPD) through in-context learning. Despite their strong empirical performance and ability to go beyond point predictions, theoretical unde…
In motion planning with partial observability, addressing uncertainty is crucial for preventing collisions and congestion, especially in the vicinity of constrained narrow areas connecting wider spaces, called hallways. In this work, we propose a cooperative motion planning algorithm that leverages
Local optima issues challenge mobile robots mapless navigation with the dilemma of avoiding collisions and approaching the target. Planning-based methods rely on environmental models and manual strategies to guide the robot. In contrast, learning-based methods can process original sensor data to nav
Transformers have emerged as powerful meta-learners, with growing evidence that they implement learning algorithms within their forward pass. We study this phenomenon in the context of denoising, presenting a unified framework that shows Transformers can implement (a) manifold denoising via Laplacia…
We show theoretically and empirically that the linear Transformer, when applied to graph data, can implement algorithms that solve canonical problems such as electric flow and eigenvector decomposition. The Transformer has access to information on the input graph only via the graph's incidence matri…
In-context learning based on attention models is examined for data with categorical outcomes, with inference in such models viewed from the perspective of functional gradient descent (GD). We develop a network composed of attention blocks, with each block employing a self-attention layer followed by…
In-Context Learning (ICL) is an essential emergent ability of Large Language Models (LLMs), and recent studies introduce CoT to exemplars of ICL to enhance the reasoning capability, especially in mathematics tasks. However, given the continuous advancement of model capabilities, it remains unclear w
We propose a continuous gathering scheme based on the swept volume to address the challenges involved in planning a tethered robot duo to efficiently collect marine debris. Specifically, we model the tethered robot duo by constructing a double-layer U-shape, and then apply an object-aware optimizati
Transformer training is notoriously difficult, requiring a careful design of optimizers and use of various heuristics. We make progress towards understanding the subtleties of training Transformers by carefully studying a simple yet canonical linearized *shallow* Transformer model. Specifically, we…
Many neural network architectures are known to be Turing Complete, and can thus, in principle implement arbitrary algorithms. However, Transformers are unique in that they can implement gradient-based learning algorithms *under simple parameter configurations*. This paper provides theoretical and em…
MCMC algorithms offer empirically efficient tools for sampling from a target distribution $\pi(x) \propto \exp(-V(x))$. However, on the theory side, MCMC algorithms suffer from slow mixing rate when $\pi(x)$ is non-log-concave. Our work examines this gap and shows that when Poincar\'e-style inequali…
State-of-the-art multi-agent collision avoidance algorithms face limitations when applied to cluttered public environments, where obstacles may have a variety of shapes and structures. The issue arises because most of these algorithms are agent-level methods. They concentrate solely on preventing co…
The Multi-Agent Path Finding (MAPF) problem is a critical challenge in dynamic multi-robot systems. Recent studies have revealed that multi-agent reinforcement learning (MARL) is a promising approach to solving MAPF problems in a fully decentralized manner. However, as the size of the multi-robot sy…
Generative processes that involve solving differential equations, such as diffusion models, frequently necessitate balancing speed and quality. ODE-based samplers are fast but plateau in performance while SDE-based samplers deliver higher sample quality at the cost of increased sampling time. We at…
Several recent works demonstrate that transformers can implement algorithms like gradient descent. By a careful construction of weights, these works show that multiple layers of transformers are expressive enough to simulate iterations of gradient descent. Going beyond the question of expressivity,…
We study the task of efficiently sampling from a Gibbs distribution $d \pi^* = e^{-h} d {\text{vol}}_g$ over a Riemannian manifold $M$ via (geometric) Langevin MCMC; this algorithm involves computing exponential maps in random Gaussian directions and is efficiently implementable in practice. The key…
Domain adaptation solves a learning problem in a target domain by utilizing the training data in a different but related source domain. As a simple and efficient method for domain adaptation, correlation alignment transforms the distribution of the source domain by utilizing the covariance matrix of…
Spiking Neural Network (SNN) is considered more biologically plausible and energy-efficient on emerging neuromorphic hardware. Recently backpropagation algorithm has been utilized for training SNN, which allows SNN to go deeper and achieve higher performance. However, most existing SNN models for ob…
We prove quantitative convergence rates at which discrete Langevin-like processes converge to the invariant distribution of a related stochastic differential equation. We study the setup where the additive noise can be non-Gaussian and state-dependent and the potential function can be non-convex. We…
We consider first order gradient methods for effectively optimizing a composite objective in the form of a sum of smooth and, potentially, non-smooth functions. We present accelerated and adaptive gradient methods, called FLAG and FLARE, which can offer the best of both worlds. They can achieve the…