Converging evidence suggests that human systems of semantic categories achieve near-optimal compression via the Information Bottleneck (IB) complexity-accuracy tradeoff. Large language models (LLMs) are not trained for this objective, which raises the question: are LLMs capable of evolving efficient human-aligned semantic systems? To address this question, we focus on color categorization -- a key testbed of cognitive theories of categorization with uniquely rich human data -- and replicate with LLMs two influential human studies. First, we conduct an English color-naming study, showing that LLMs vary widely in their complexity and English-alignment, with larger instruction-tuned models achieving better alignment and IB-efficiency. Second, to test whether these LLMs simply mimic patterns in their training data or actually exhibit a human-like inductive bias toward IB-efficiency, we simulate cultural evolution of pseudo color-naming systems in LLMs via a method we refer to as Iterated in-Context Language Learning (IICLL). We find that akin to humans, LLMs iteratively restructure initially random systems towards greater IB-efficiency. However, only a model with strongest in-context capabilities (Gemini 2.0) is able to recapitulate the wide range of near-optimal IB-tradeoffs observed in humans, while other state-of-the-art models converge to low-complexity solutions. These findings demonstrate how human-aligned semantic categories can emerge in LLMs via the same fundamental principle that underlies semantic efficiency in humans.

Hand gesture classification using high-quality structured data such as videos, images, and hand skeletons is a well-explored problem in computer vision. Leveraging low-power, cost-effective biosignals, e.g. surface electromyography (sEMG), allows for continuous gesture prediction on wearables. In this paper, we demonstrate that learning representations from weak-modality data that are aligned with those from structured, high-quality data can improve representation quality and enables zero-shot classification. Specifically, we propose a Contrastive Pose-EMG Pre-training (CPEP) framework to align EMG and pose representations, where we learn an EMG encoder that produces high-quality and pose-informative representations. We assess the gesture classification performance of our model through linear probing and zero-shot setups. Our model outperforms emg2pose benchmark models by up to 21% on in-distribution gesture classification and 72% on unseen (out-of-distribution) gesture classification.

Many of language models' impressive capabilities originate from their in-context learning: based on instructions or examples, they can infer and perform new tasks without weight updates. In this work, we investigate when representations for new tasks are formed in language models, and how these representations change over the course of context. We study two different task representations: those that are ''transferrable'' -- vector representations that can transfer task contexts to another model instance, even without the full prompt -- and simpler representations of high-level task categories. We show that transferrable task representations evolve in non-monotonic and sporadic ways, while task identity representations persist throughout the context. Specifically, transferrable task representations exhibit a two-fold locality. They successfully condense evidence when more examples are provided in the context. But this evidence accrual process exhibits strong temporal locality along the sequence dimension, coming online only at certain tokens -- despite task identity being reliably decodable throughout the context. In some cases, transferrable task representations also show semantic locality, capturing a small task ''scope'' such as an independent subtask. Language models thus represent new tasks on the fly through both an inert, sustained sensitivity to the task and an active, just-in-time representation to support inference.

Fine-tuning large language models (LLMs) is crucial for adapting them to specific tasks, yet it remains computationally demanding and raises concerns about correctness and privacy, particularly in untrusted environments. Although parameter-efficient methods like Low-Rank Adaptation (LoRA) significantly reduce resource requirements, ensuring the security and verifiability of fine-tuning under zero-knowledge constraints remains an unresolved challenge. To address this, we introduce VeriLoRA, the first framework to integrate LoRA fine-tuning with zero-knowledge proofs (ZKPs), achieving provable security and correctness. VeriLoRA employs advanced cryptographic techniques -- such as lookup arguments, sumcheck protocols, and polynomial commitments -- to verify both arithmetic and non-arithmetic operations in Transformer-based architectures. The framework provides end-to-end verifiability for forward propagation, backward propagation, and parameter updates during LoRA fine-tuning, while safeguarding the privacy of model parameters and training data. Leveraging GPU-based implementations, VeriLoRA demonstrates practicality and efficiency through experimental validation on open-source LLMs like LLaMA, scaling up to 13 billion parameters. By combining parameter-efficient fine-tuning with ZKPs, VeriLoRA bridges a critical gap, enabling secure and trustworthy deployment of LLMs in sensitive or untrusted environments.

This paper introduces a novel research direction for model-to-text/code transformations by leveraging Large Language Models (LLMs) that can be enhanced with Retrieval-Augmented Generation (RAG) pipelines. The focus is on quantum and hybrid quantum-classical software systems, where model-driven approaches can help reduce the costs and mitigate the risks associated with the heterogeneous platform landscape and lack of developers' skills. We validate one of the proposed ideas regarding generating code out of UML model instances of software systems. This Python code uses a well-established library, called Qiskit, to execute on gate-based or circuit-based quantum computers. The RAG pipeline that we deploy incorporates sample Qiskit code from public GitHub repositories. Experimental results show that well-engineered prompts can improve CodeBLEU scores by up to a factor of four, yielding more accurate and consistent quantum code. However, the proposed research direction can go beyond this through further investigation in the future by conducting experiments to address our other research questions and ideas proposed here, such as deploying software system model instances as the source of information in the RAG pipelines, or deploying LLMs for code-to-code transformations, for instance, for transpilation use cases.

GRAFT is a structured multimodal benchmark designed to probe how well LLMs handle instruction following, visual reasoning, and tasks requiring tight visual textual alignment. The dataset is built around programmatically generated charts and synthetically rendered tables, each paired with a carefully constructed, multi step analytical question that depends solely on what can be inferred from the image itself. Responses are formatted in structured outputs such as JSON or YAML, enabling consistent and fine grained evaluation of both reasoning processes and adherence to output specifications. The benchmark further introduces a taxonomy of reasoning operations ranging from comparison and trend identification to ranking, aggregation, proportional estimation, and anomaly detection to support a comprehensive assessment of model capabilities. Taken together, GRAFT provides a unified and scalable framework for evaluating multimodal LLMs on visually grounded, structured reasoning tasks, offering a more rigorous standard for future benchmarking efforts.

LLM-based web agents have the potential to automate long-running web tasks, such as searching for products in multiple e-shops and subsequently ordering the cheapest products that meet the users needs. Benchmarks for evaluating web agents either require agents to perform tasks online using the live Web or offline using simulated environments, which allow for the exact reproduction of the experimental setup. While DeepShop provides an online benchmark that requires agents to perform challenging shopping tasks, existing offline benchmarks such as WebShop, WebArena, or Mind2Web cover only comparatively simple e-commerce tasks that need to be performed against a single shop containing product data from a single source. What is missing is an e-commerce benchmark that simulates multiple shops containing heterogeneous product data and requires agents to perform complex tasks. We fill this gap by introducing WebMall, the first offline multi-shop benchmark for evaluating web agents on challenging comparison shopping tasks. WebMall consists of four simulated shops populated with product data extracted from the Common Crawl. The WebMall tasks range from specific product searches and price comparisons to advanced queries for complementary or substitute products, as well as checkout processes. We validate WebMall using eight agents that differ in observation space, availability of short-term memory, and the employed LLM. The validation highlights the difficulty of the benchmark, with even the best-performing agents achieving task completion rates below 55% in the task categories cheapest product search and vague product search.

To investigate the feasibility of zero-shot self-supervised learning reconstruction for reducing breath-hold times in magnetic resonance cholangiopancreatography (MRCP). Breath-hold MRCP was acquired from 11 healthy volunteers on 3T scanners using an incoherent k-space sampling pattern, leading to 14-second acquisition time and an acceleration factor of R=25. Zero-shot reconstruction was compared with parallel imaging of respiratory-triggered MRCP (338s, R=3) and compressed sensing reconstruction. For two volunteers, breath-hold scans (40s, R=6) were additionally acquired and retrospectively undersampled to R=25 to compute peak signal-to-noise ratio (PSNR). To address long zero-shot training time, the n+m full stages of the zero-shot learning were divided into two parts to reduce backpropagation depth during training: 1) n frozen stages initialized with n-stage pretrained network and 2) m trainable stages initialized either randomly or m-stage pretrained network. Efficiency of our approach was assessed by varying initialization strategies and the number of trainable stages using the retrospectively undersampled data. Zero-shot reconstruction significantly improved visual image quality over compressed sensing, particularly in SNR and ductal delineation, and achieved image quality comparable to that of successful respiratory-triggered acquisitions with regular breathing patterns. Improved initializations enhanced PSNR and reduced reconstruction time. Adjusting frozen/trainable configurations demonstrated that PSNR decreased only slightly from 38.25 dB (0/13) to 37.67 dB (12/1), while training time decreased up to 6.7-fold. Zero-shot learning delivers high-fidelity MRCP reconstructions with reduced breath-hold times, and the proposed partially trainable approach offers a practical solution for translation into time-constrained clinical workflows.

Although large language models (LLMs) hold significant promise in psychotherapy, their direct application in patient-facing scenarios raises ethical and safety concerns. Therefore, this work shifts towards developing an LLM as a supervisor to train real therapists. In addition to the privacy of clinical therapist training data, a fundamental contradiction complicates the training of therapeutic behaviors: clear feedback standards are necessary to ensure a controlled training system, yet there is no absolute "gold standard" for appropriate therapeutic behaviors in practice. In contrast, many common therapeutic mistakes are universal and identifiable, making them effective triggers for targeted feedback that can serve as clearer evidence. Motivated by this, we create a novel therapist-training paradigm: (1) guidelines for mistaken behaviors and targeted correction strategies are first established as standards; (2) a human-in-the-loop dialogue-feedback dataset is then constructed, where a mistake-prone agent intentionally makes standard mistakes during interviews naturally, and a supervisor agent locates and identifies mistakes and provides targeted feedback; (3) after fine-tuning on this dataset, the final supervisor model is provided for real therapist training. The detailed experimental results of automated, human and downstream assessments demonstrate that models fine-tuned on our dataset MATE, can provide high-quality feedback according to the clinical guideline, showing significant potential for the therapist training scenario.

Abstract--Data-intensive Artificial Intelligence (AI) applications at the network edge demand a fundamental shift in Radio Access Network (RAN) design, from merely consuming AI for network optimization, to actively enabling distributed AI workloads. This presents a significant opportunity for network operators to monetize AI while leveraging existing infrastructure. T o realize this vision, this article presents a novel converged O-RAN and AI-RAN architecture for unified orchestration and management of telecommunications and AI workloads on shared infrastructure. The proposed architecture extends the Open RAN principles of modularity, disaggregation, and cloud-nativeness to support heterogeneous AI deployments. We introduce two key architectural innovations: (i) the AI-RAN Orchestrator, which extends the O-RAN Service Management and Orchestration (SMO) to enable integrated resource and allocation across RAN and AI workloads; and (ii) AI-RAN sites that provide distributed edge AI platforms with real-time processing capabilities. The proposed architecture enables flexible orchestration, meeting requirements for managing heterogeneous workloads at different time scales while maintaining open, standardized interfaces and multi-vendor interoperability.This paper has been submitted to IEEE for publication. M. Polese, L. Bonati, and T. Melodia are with the Institute for the Wireless Internet of Things, Northeastern University, Boston, MA, USA. This article is based upon work partially supported by the NTIA PWSCIF under A ward No. 25-60-IF054, the U.S. NSF under award CNS-2112471, and by OUSD(R&E) through Army Research Laboratory Cooperative Agreement Number W911NF-24-2-0065.