Eliciting Better Multilingual Structured Reasoning from LLMs through Code

Development of large language models (LLM) have shown progress on reasoning, though studies have been limited to English or simple reasoning tasks. We thus introduce a multilingual structured reasoning and explanation dataset, termed xSTREET, that covers four tasks across six languages. xSTREET exposes a gap in base LLM performance between English and non-English reasoning tasks. We then propose two methods to remedy this gap, building on the insight that LLMs trained on code are better reasoners. First, at training time, we augment a code dataset with multi-lingual comments using machine translation while keeping program code as-is. Second, at inference time, we bridge the gap between training and inference by employing a prompt structure that incorporates step-by-step code primitives to derive new facts and find a solution. Our methods show improved multilingual performance on xSTREET, most notably on the scientific commonsense reasoning subtask. Furthermore, the models show no regression on non-reasoning tasks, thus showing our techniques maintain general-purpose abilities.

MAGID: An Automated Pipeline for Generating Synthetic Multi-modal Datasets

Development of multimodal interactive systems is hindered by the lack of rich, multimodal (text, images) conversational data, which is needed in large quantities for LLMs. Previous approaches augment textual dialogues with retrieved images, posing privacy, diversity, and quality constraints. In this work, we introduce \textbf{M}ultimodal \textbf{A}ugmented \textbf{G}enerative \textbf{I}mages \textbf{D}ialogues (MAGID), a framework to augment text-only dialogues with diverse and high-quality images. Subsequently, a diffusion model is applied to craft corresponding images, ensuring alignment with the identified text. Finally, MAGID incorporates an innovative feedback loop between an image description generation module (textual LLM) and image quality modules (addressing aesthetics, image-text matching, and safety), that work in tandem to generate high-quality and multi-modal dialogues. We compare MAGID to other SOTA baselines on three dialogue datasets, using automated and human evaluation. Our results show that MAGID is comparable to or better than baselines, with significant improvements in human evaluation, especially against retrieval baselines where the image database is small.

Semantic Text Transmission via Prediction with Small Language Models: Cost-Similarity Trade-off

We consider the communication of natural language text from a source to a destination over noiseless and character-erasure channels. We exploit language's inherent correlations and predictability to constrain transmission costs by allowing the destination to predict or complete words with potential dissimilarity with the source text. Concretely, our objective is to obtain achievable $(\bar{c}, \bar{s})$ pairs, where $\bar{c}$ is the average transmission cost at the source and $\bar{s}$ is the average semantic similarity measured via cosine similarity between vector embedding of words at the source and those predicted/completed at the destination. We obtain $(\bar{c}, \bar{s})$ pairs for neural language and first-order Markov chain-based small language models (SLM) for prediction, using both a threshold policy that transmits a word if its cosine similarity with that predicted/completed at the destination is below a threshold, and a periodic policy, which transmits words after a specific interval and predicts/completes the words in between, at the destination. We adopt an SLM for word completion. We demonstrate that, when communication occurs over a noiseless channel, the threshold policy achieves a higher $\bar{s}$ for a given $\bar{c}$ than the periodic policy and that the $\bar{s}$ achieved with the neural SLM is greater than or equal to that of the Markov chain-based algorithm for the same $\bar{c}$. The improved performance comes with a higher complexity in terms of time and computing requirements. However, when communication occurs over a character-erasure channel, all prediction algorithms and scheduling policies perform poorly. Furthermore, if character-level Huffman coding is used, the required $\bar{c}$ to achieve a given $\bar{s}$ is reduced, but the above observations still apply.

Version age-based client scheduling policy for federated learning

Federated Learning (FL) has emerged as a privacy-preserving machine learning paradigm facilitating collaborative training across multiple clients without sharing local data. Despite advancements in edge device capabilities, communication bottlenecks present challenges in aggregating a large number of clients; only a portion of the clients can update their parameters upon each global aggregation. This phenomenon introduces the critical challenge of stragglers in FL and the profound impact of client scheduling policies on global model convergence and stability. Existing scheduling strategies address staleness but predominantly focus on either timeliness or content. Motivated by this, we introduce the novel concept of Version Age of Information (VAoI) to FL. Unlike traditional Age of Information metrics, VAoI considers both timeliness and content staleness. Each client's version age is updated discretely, indicating the freshness of information. VAoI is incorporated into the client scheduling policy to minimize the average VAoI, mitigating the impact of outdated local updates and enhancing the stability of FL systems.

DeAL: Decoding-time Alignment for Large Language Models

Large Language Models (LLMs) are nowadays expected to generate content aligned with human preferences. Current work focuses on alignment at model training time, through techniques such as Reinforcement Learning with Human Feedback (RLHF). However, it is unclear if such methods are an effective choice to teach alignment objectives to the model. First, the inability to incorporate multiple, custom rewards and reliance on a model developer's view of universal and static principles are key limitations. Second, the residual gaps in model training and the reliability of such approaches are also questionable (e.g. susceptibility to jail-breaking even after safety training). To address these, we propose DeAL, a framework that allows the user to customize reward functions and enables Decoding-time Alignment of LLMs (DeAL). At its core, we view decoding as a heuristic-guided search process and facilitate the use of a wide variety of alignment objectives. Our experiments with programmatic constraints such as keyword and length constraints (studied widely in the pre-LLM era) and abstract objectives such as harmlessness and helpfulness (proposed in the post-LLM era) show that we can DeAL with fine-grained trade-offs, improve adherence to alignment objectives, and address residual gaps in LLMs. Lastly, while DeAL can be effectively paired with RLHF and prompting techniques, its generality makes decoding slower, an optimization we leave for future work.

Goal-Oriented Communication, Estimation, and Control over Bidirectional Wireless Links

We consider a wireless networked control system (WNCS) with bidirectional imperfect links for real-time applications such as smart grids. To maintain the stability of WNCS, captured by the probability that plant state violates preset values, at minimal cost, heterogeneous physical processes are monitored by multiple sensors. This status information, such as dynamic plant state and Markov Process-based context information, is then received/estimated by the controller for remote control. However, scheduling multiple sensors and designing the controller with limited resources is challenging due to their coupling, delay, and transmission loss. We formulate a Constrained Markov Decision Problem (CMDP) to minimize violation probability with cost constraints. We reveal the relationship between the goal and different updating actions by analyzing the significance of information that incorporates goal-related usefulness and contextual importance. Subsequently, a goal-oriented deterministic scheduling policy is proposed. Two sensing-assisted control strategies and a control-aware estimation policy are proposed to improve the violation probability-cost tradeoff, integrated with the scheduling policy to form a goal-oriented co-design framework. Additionally, we explore retransmission in downlink transmission and qualitatively analyze its preference scenario. Simulation results demonstrate that the proposed goal-oriented co-design policy outperforms previous work in simultaneously reducing violation probability and cost

Age of Actuation and Timeliness: Semantics in a Wireless Power Transfer System

In this paper, we investigate a model relevant to semantics-aware goal-oriented communications, and we propose a new metric that incorporates the utilization of information in addition to its timelines. Specifically, we consider the transmission of observations from an external process to a battery-powered receiver through status updates. These updates inform the receiver about the process status and enable actuation if sufficient energy is available to achieve a goal. We focus on a wireless power transfer (WPT) model, where the receiver receives energy from a dedicated power transmitter and occasionally from the data transmitter when they share a common channel. We analyze the Age of Information (AoI) and propose a new metric, the \textit{Age of Actuation (AoA), which is relevant when the receiver utilizes the status updates to perform actions in a timely manner}. We provide analytical characterizations of the average AoA and the violation probability of the AoA, demonstrating that AoA generalizes AoI. Moreover, we introduce and analytically characterize the \textit{Probability of Missing Actuation (PoMA)}; this metric becomes relevant also \textit{to quantify the incurred cost of a missed action}. We formulate unconstrained and constrained optimization problems for all the metrics and present numerical evaluations of our analytical results. This proposed set of metrics goes beyond the traditional timeliness metrics since the synergy of different flows is now considered.

Value of Information and Timing-aware Scheduling for Federated Learning

Data possesses significant value as it fuels advancements in AI. However, protecting the privacy of the data generated by end-user devices has become crucial. Federated Learning (FL) offers a solution by preserving data privacy during training. FL brings the model directly to User Equipments (UEs) for local training by an access point (AP). The AP periodically aggregates trained parameters from UEs, enhancing the model and sending it back to them. However, due to communication constraints, only a subset of UEs can update parameters during each global aggregation. Consequently, developing innovative scheduling algorithms is vital to enable complete FL implementation and enhance FL convergence. In this paper, we present a scheduling policy combining Age of Update (AoU) concepts and data Shapley metrics. This policy considers the freshness and value of received parameter updates from individual data sources and real-time channel conditions to enhance FL's operational efficiency. The proposed algorithm is simple, and its effectiveness is demonstrated through simulations.

Optimal Status Updates for Minimizing Age of Correlated Information in IoT Networks with Energy Harvesting Sensors

Many real-time applications of the Internet of Things (IoT) need to deal with correlated information generated by multiple sensors. The design of efficient status update strategies that minimize the Age of Correlated Information (AoCI) is a key factor. In this paper, we consider an IoT network consisting of sensors equipped with the energy harvesting (EH) capability. We optimize the average AoCI at the data fusion center (DFC) by appropriately managing the energy harvested by sensors, whose true battery states are unobservable during the decision-making process. Particularly, we first formulate the dynamic status update procedure as a partially observable Markov decision process (POMDP), where the environmental dynamics are unknown to the DFC. In order to address the challenges arising from the causality of energy usage, unknown environmental dynamics, unobservability of sensors'true battery states, and large-scale discrete action space, we devise a deep reinforcement learning (DRL)-based dynamic status update algorithm. The algorithm leverages the advantages of the soft actor-critic and long short-term memory techniques. Meanwhile, it incorporates our proposed action decomposition and mapping mechanism. Extensive simulations are conducted to validate the effectiveness of our proposed algorithm by comparing it with available DRL algorithms for POMDPs.

Pre-training Intent-Aware Encoders for Zero- and Few-Shot Intent Classification

Intent classification (IC) plays an important role in task-oriented dialogue systems as it identifies user intents from given utterances. However, models trained on limited annotations for IC often suffer from a lack of generalization to unseen intent classes. We propose a novel pre-training method for text encoders that uses contrastive learning with intent psuedo-labels to produce embeddings that are well-suited for IC tasks. By applying this pre-training strategy, we also introduce the pre-trained intent-aware encoder (PIE). Specifically, we first train a tagger to identify key phrases within utterances that are crucial for interpreting intents. We then use these extracted phrases to create examples for pre-training a text encoder in a contrastive manner. As a result, our PIE model achieves up to 5.4% and 4.0% higher accuracy than the previous state-of-the-art pre-trained sentence encoder for the N-way zero- and one-shot settings on four IC datasets.