Protecting the Trace: A Principled Black-Box Approach Against Distillation Attacks

Frontier models push the boundaries of what is learnable at extreme computational costs, yet distillation via sampling reasoning traces exposes closed-source frontier models to adversarial third parties who can bypass their guardrails and misappropriate their capabilities, raising safety, security, and intellectual privacy concerns. To address this, there is growing interest in building antidistillation methods, which aim to poison reasoning traces to hinder downstream student model learning while maintaining teacher performance. However, current techniques lack theoretical grounding, requiring either heavy fine-tuning or access to student model proxies for gradient based attacks, and often lead to a significant teacher performance degradation. In this work, we present a theoretical formulation of antidistillation as a Stackelberg game, grounding a problem that has so far largely been approached heuristically. Guided by the desired design properties our formulation reveals, we propose \texttt{TraceGuard}, an efficient, post-generation black-box method to poison sentences with high importance for teacher reasoning. Our work offers a scalable solution to share model insights safely, ensuring that the advancement of reasoning capabilities does not come at the cost of intellectual privacy or AI safety alignment.

Energy-Aware Routing to Large Reasoning Models

Large reasoning models (LRMs) have heterogeneous inference energy costs based on which model is used and how much it reasons. To reduce energy, it is important to choose the right LRM and operate it in the right way. As a result, the performance of systems that dispatch tasks to different individual LRMs depend on the balance between mean energy provisioning and stochastic fluctuations. The critical regime is the unique operating point at which neither auxiliary energy nor baseline energy is systematically wasted. Increasing baseline supply shifts the system toward persistent over-supply and baseline-energy waste, while reducing supply induces persistent reliance on auxiliary energy. Yet in this regime, performance remains volatility-limited and so a second-order characterization provides further insights that we develop. Here, performance is governed by how variability is absorbed across time, models, and execution choices. This perspective highlights variance-aware routing and dispatch as a principled design axis, and provides a theoretical basis for developing energy-aware model routing policies. Routing behavior is characterized when dispatch policies are based on training-compute and inference-compute scaling laws for LRMs.

Nonlinear Federated System Identification

We consider federated learning of linearly-parameterized nonlinear systems. We establish theoretical guarantees on the effectiveness of federated nonlinear system identification compared to centralized approaches, demonstrating that the convergence rate improves as the number of clients increases. Although the convergence rates in the linear and nonlinear cases differ only by a constant, this constant depends on the feature map $\phi$, which can be carefully chosen in the nonlinear setting to increase excitation and improve performance. We experimentally validate our theory in physical settings where client devices are driven by i.i.d. control inputs and control policies exhibiting i.i.d. random perturbations, ensuring non-active exploration. Experiments use trajectories from nonlinear dynamical systems characterized by real-analytic feature functions, including polynomial and trigonometric components, representative of physical systems including pendulum and quadrotor dynamics. We analyze the convergence behavior of the proposed method under varying noise levels and data distributions. Results show that federated learning consistently improves convergence of any individual client as the number of participating clients increases.

SparseJEPA: Sparse Representation Learning of Joint Embedding Predictive Architectures

Joint Embedding Predictive Architectures (JEPA) have emerged as a powerful framework for learning general-purpose representations. However, these models often lack interpretability and suffer from inefficiencies due to dense embedding representations. We propose SparseJEPA, an extension that integrates sparse representation learning into the JEPA framework to enhance the quality of learned representations. SparseJEPA employs a penalty method that encourages latent space variables to be shared among data features with strong semantic relationships, while maintaining predictive performance. We demonstrate the effectiveness of SparseJEPA by training on the CIFAR-100 dataset and pre-training a lightweight Vision Transformer. The improved embeddings are utilized in linear-probe transfer learning for both image classification and low-level tasks, showcasing the architecture's versatility across different transfer tasks. Furthermore, we provide a theoretical proof that demonstrates that the grouping mechanism enhances representation quality. This was done by displaying that grouping reduces Multiinformation among latent-variables, including proofing the Data Processing Inequality for Multiinformation. Our results indicate that incorporating sparsity not only refines the latent space but also facilitates the learning of more meaningful and interpretable representations. In further work, hope to further extend this method by finding new ways to leverage the grouping mechanism through object-centric representation learning.

SwitchCIT: Switching for Continual Instruction Tuning of Large Language Models

Large language models (LLMs) have exhibited impressive capabilities in various domains, particularly in general language understanding. However these models, trained on massive text data, may not be finely optimized for specific tasks triggered by instructions. Continual instruction tuning is crucial to adapt LLMs to evolving tasks and domains, ensuring their effectiveness and relevance across a wide range of applications. In the context of continual instruction tuning, where models are sequentially trained on different tasks, catastrophic forgetting can occur, leading to performance degradation on previously learned tasks. This work addresses the catastrophic forgetting in continual instruction learning for LLMs through a switching mechanism for routing computations to parameter-efficient tuned models. We demonstrate the effectiveness of our method through experiments on continual instruction tuning of different natural language generation tasks.