Argyrios Gerogiannis

Yekaterina Yegorova, Argyrios Gerogiannis, Haolong Zheng et al.

Speech-aware large language models often generalize poorly to out-of-domain settings. We propose SALSA (Speech-Aware LLM Adaptation via Learned Steering Activations), a lightweight adaptation method that learns layer-wise steering vectors. Unlike commonly used steering approaches that rely on contrastive activation differences, SALSA directly optimizes steering vectors using a supervised objective. Across children's speech, multilingual speech, and Mandarin-English code-switching benchmarks, SALSA substantially improves performance over zero-shot inference and speech in-context learning baselines, achieving up to 46.8% relative improvements over zero-shot. Analysis further demonstrates that steering the encoder, particularly the later layers, is more effective than steering the LLM backbone. These findings suggest that steering improves downstream ASR performance by adapting higher-level acoustic and phonetic representations to better align with the pretrained language model representation space, rather than by modifying the decoder itself.

Yu-Han Huang, Argyrios Gerogiannis, Subhonmesh Bose et al.

We study the multichannel quickest change detection problem with bandit feedback and controlled sensing, in which an agent sequentially selects one of the data streams to observe at each time-step and aims to detect an unknown change as quickly as possible while controlling false alarms. Assuming known pre- and post-change distributions and allowing an arbitrary subset of streams to be affected by the change, we propose two novel and computationally efficient detection procedures inspired by the Upper Confidence Bound (UCB) multi-armed bandit algorithm. Our methods adaptively concentrate sensing on the most informative streams while preserving false-alarm guarantees. We show that both procedures achieve first-order asymptotic optimality in detection delay under standard false-alarm constraints. We also extend the UCB-driven controlled sensing approach to the setting where the pre- and post-change distributions are unknown, except for a mean-shift in at least one of the channels at the change-point. This setting is particularly relevant to the problem of learning in piecewise stationary environments. Finally, extensive simulations on synthetic benchmarks show that our methods consistently outperform existing state-of-the-art approaches while offering substantial computational savings.

Argyrios Gerogiannis, Yu-Han Huang, Venugopal V. Veeravalli

We study model-free reinforcement learning (RL) in non-stationary finite-horizon episodic Markov decision processes (MDPs) without prior knowledge of the non-stationarity. We focus on the piecewise-stationary (PS) setting, where both the reward and transition dynamics can change an arbitrary number of times. We propose Detection Augmented Reinforcement Learning (DARLING), a modular wrapper for PS-RL that applies to both tabular and linear MDPs, without knowledge of the changes. Under certain change-point separation and reachability conditions, DARLING improves the best available dynamic regret bounds in both settings and yields strong empirical performance. We further establish the first minimax lower bounds for PS-RL in tabular and linear MDPs, showing that DARLING is the first nearly optimal algorithm. Experiments on standard benchmarks demonstrate that DARLING consistently surpasses the state-of-the-art methods across diverse non-stationary scenarios.

Moulik Choraria, Argyrios Gerogiannis, Anirban Das et al.

Compute scaling for LLM reasoning requires allocating budget between exploring solution approaches ($breadth$) and refining promising solutions ($depth$). Most methods implicitly trade off one for the other, yet why a given trade-off works remains unclear, and validation on a single model obscures the role of the model itself. We argue that $\textbf{the optimal strategy depends on the model's diversity profile, the spread of probability mass across solution approaches, and that this must be characterized before any exploration strategy is adopted.}$ We formalize this through a theoretical framework decomposing reasoning uncertainty and derive conditions under which tree-style depth refinement outperforms parallel sampling. We validate it on Qwen-3 4B and Olmo-3 7B families, showing that lightweight signals suffice for depth-based refinement on low-diversity aligned models while yielding limited utility for high-diversity base models, which we hypothesize require stronger compensation for lower exploration coverage.

Moulik Choraria, Argyrios Gerogiannis, Vidhata Jayaraman et al.

Hopfield networks and their generalizations have established deep connections among biological associative memories, statistical physics, and transformers. Yet most models treat retrieval as a fixed query-to-memory mapping, ignoring the role of external context in recall. In this work, we propose a two-stage associative memory architecture, wherein a context-gate subcircuit reshapes the retrieval energy landscape before and during recall. We show theoretically that context gating increases inter-memory separation while inducing sparsity, translating into exponential improvements in retrieval. Crucially, we prove that the system admits a unique self-consistent fixed point, revealing that the resulting retrieval state is driven by both a direct contextual bias and a second-order retrieval-gate feedback loop. We then bridge this theory to transformers; specifically, we evaluate a first-order approximation on Llama-3, confirming that in-context learning acts as context-gated retrieval. Native dynamics mirror our theory: context localizes a memory subspace, enabling the zero-shot query to cleanly discriminate. Ultimately, this framework provides a mechanistic link between associative memory theory and LLM phenomenology.

Argyrios Gerogiannis, Yu-Han Huang, Venugopal V. Veeravalli

We study the problem of Non-Stationary Reinforcement Learning (NS-RL) without prior knowledge about the system's non-stationarity. A state-of-the-art, black-box algorithm, known as MASTER, is considered, with a focus on identifying the conditions under which it can achieve its stated goals. Specifically, we prove that MASTER's non-stationarity detection mechanism is not triggered for practical choices of horizon, leading to performance akin to a random restarting algorithm. Moreover, we show that the regret bound for MASTER, while being order optimal, stays above the worst-case linear regret until unreasonably large values of the horizon. To validate these observations, MASTER is tested for the special case of piecewise stationary multi-armed bandits, along with methods that employ random restarting, and others that use quickest change detection to restart. A simple, order optimal random restarting algorithm, that has prior knowledge of the non-stationarity is proposed as a baseline. The behavior of the MASTER algorithm is validated in simulations, and it is shown that methods employing quickest change detection are more robust and consistently outperform MASTER and other random restarting approaches.

Yu-Han Huang, Argyrios Gerogiannis, Subhonmesh Bose et al.

Conventional Multi-Armed Bandit (MAB) algorithms are designed for stationary environments, where the reward distributions associated with the arms do not change with time. In many applications, however, the environment is more accurately modeled as being non-stationary. In this work, piecewise stationary MAB (PS-MAB) environments are investigated, in which the reward distributions associated with a subset of the arms change at some change-points and remain stationary between change-points. Our focus is on the asymptotic analysis of PS-MABs, for which practical algorithms based on change detection have been previously proposed. Our goal is to modularize the design and analysis of such Detection Augmented Bandit (DAB) procedures. To this end, we first provide novel, improved performance lower bounds for PS-MABs. Then, we identify the requirements for stationary bandit algorithms and change detectors in a DAB procedure that are needed for the modularization. We assume that the rewards are sub-Gaussian. Under this assumption and a condition on the separation of the change-points, we show that the analysis of DAB procedures can indeed be modularized, so that the regret bounds can be obtained in a unified manner for various combinations of change detectors and bandit algorithms. Through this analysis, we develop new modular DAB procedures that are order-optimal. Finally, we showcase the practical effectiveness of our modular DAB approach in our experiments, studying its regret performance compared to other methods and investigating its detection capabilities.

Argyrios Gerogiannis, Yu-Han Huang, Subhonmesh Bose et al.

We introduce a practical, black-box framework termed Detection Augmented Learning (DAL) for the problem of non-stationary bandits without prior knowledge of the underlying non-stationarity. DAL accepts any stationary bandit algorithm as input and augments it with a change detector, enabling applicability to all common bandit variants. Extensive experimentation demonstrates that DAL consistently surpasses current state-of-the-art methods across diverse non-stationary scenarios, including synthetic benchmarks and real-world datasets, underscoring its versatility and scalability. We provide theoretical insights into DAL's strong empirical performance, complemented by thorough experimental validation.