Computer Science & Mathematics

Regularised reinforcement learning (RL) has provable robustness guarantees that should ease the transfer of policies from simulation to robots, known as sim-to-real. Yet, weakly regularised algorithms like Proximal Policy Optimisation (PPO) remain the de facto choice in sim-to-real pipelines, while the practical benefits of regularisation for transfer remain poorly characterised. In this work, we study the sim-to-real performance of commonly used RL algorithms, with the aim of characterising the effects of regularisation on robotic performance. To that end, we derive Soft MellowMax Actor-Critic (SMM-AC), a novel regularised RL algorithm based on the soft mellowmax value function, which carries strong robustness guarantees. On MuJoCo continuous control benchmarks, SMM-AC is competitive with baselines like Soft Actor-Critic (SAC). On real robots, SMM-AC achieves higher return, while exhibiting greater robustness to perturbations in observations, the environment, and dynamics than baselines. Together, these results demonstrate that regularised RL improves real-world policy deployment through better robustness to the sim-to-real gap, and changing real-world conditions.

Smartphone use while walking reduces environmental awareness and can lead to accidents and injury, yet existing obstacle-awareness systems remain unadopted because they require users to hold the phone upright and aimed forward, conflicting with how people actually hold their devices. We introduce GlanceAR, an always-on mobile obstacle-awareness system designed for natural phone-holding postures. GlanceAR utilizes a low-cost passive mirror attachment to redirect the phone's rear LiDAR and RGB camera forward without additional electronic components. In a within-subjects study with N=11 participants navigating indoor and outdoor obstacle courses while engaged in a primary on-phone task, we evaluated three sensing configurations: GlanceAR, a LiDAR-only variant for privacy, and an attachment-free ultra-wide camera baseline. Our findings reveal trade-offs among sensing fidelity, obstacle cue design, and social acceptability, and surface insights around AI complacency and the tension between on-screen task focus and physical safety.

Time Series Foundation Models (TSFMs) offer strong zero-shot forecasting capabilities, but their high pretraining cost raises a practical sustainability question: does this upfront investment pay off over a full deployment lifecycle, compared to supervised standalone models that must be periodically retrained? We model the cumulative carbon cost of different approaches across a realistic industrial scenario: Chronos-2 amortizes its pretraining cost within months for large or frequently queried deployments, but a classical model remains cheaper indefinitely for small or rarely queried ones. For cases where reducing the TSFM's own cost matters, we show that removing a single encoder block (7.9% of parameters) via a gradient-free pruning criterion yields a consistent 6 to 9% reduction in FLOPs, CO2 emissions, and latency, at essentially no cost to zero-shot forecasting performance across a 25-task benchmark.

Transmission expansion planning (TEP) seeks to identify cost-effective investment strategies that enable power systems to reliably meet future demand, while accounting for the location, timing, and operational impacts of infrastructure upgrades. As demand grows and renewable penetration increases, there is growing interest in jointly modeling energy storage investments to enhance operational flexibility. This work presents a multiperiod formulation that co-optimizes transmission reconductoring and energy storage installation across multiple planning periods. To solve the resulting large-scale problem efficiently, a Benders decomposition with per-period trust-region stabilization and a relaxation-based warm-start procedure is proposed. The approach is demonstrated on a 2,000-bus Texas system, producing an investment plan spanning 2030-2045 with optimality gap below 1.0%, and is compared against a rolling-horizon approach to highlight the benefits of multiperiod coordination.

Longitudinal causal effect estimation is complicated by time-varying confounding and treatment-confounder feedback loops. Existing methods generally require data-specific nuisance-model fitting, and many are tailored to prespecified treatment sequences or prediction horizons, leading to time-consuming computation at deployment. We introduce *Policy In-Context Longitudinal causal Effect estimation with Prior-data Fitted Networks* (**PICLE-PFN**), the first foundation model for longitudinal causal effect estimation that supports *dynamic* treatment policies. PICLE-PFN represents target policies through in-context history-decision demonstrations, providing a unified interface for a broad class of treatment policies. Pretrained over synthetic data sampled from longitudinal structural causal models, PICLE-PFN amortizes causal effect estimation across datasets, policies, and horizons. Given a new observational dataset and target policy, it directly predicts conditional average potential outcomes (CAPOs) in one forward pass without task-specific training or hyperparameter tuning. Across synthetic and semi-synthetic longitudinal benchmarks, PICLE-PFN achieves competitive RMSE and substantially improved uncertainty calibration with a fraction of deployment runtime, compared to state-of-the-art conventional estimators trained and tuned separately for each dataset.

This work introduces PoKER, a novel probabilistic model engineered to optimize missile launch effectiveness in air-to-air scenarios, specifically within Beyond Visual Range (BVR) air combat. Unlike conventional Weapon Engagement Zone (WEZ) models that delineate zones based on static distances such as maximum, minimum, and no-escape ranges, PoKER applies machine learning to predict kill probabilities more accurately by integrating the stochastic behaviors of targets and missile miss distances. This model dynamically evaluates target behavior, greatly expanding the predictive capabilities of engagement analysis. By factoring in elements such as warhead lethality, target and shooter orientations, and the specific conditions of engagement, PoKER provides important insights into engagement dynamics and quantifies success probabilities. Consequently, it can potentially be an important tool for BVR air combat pilots, improving operational decision-making within this specialized combat domain.

Fine-tuning on datasets of unknown origin can create a backdoor in the trained model when the data includes poisoned examples. Difficult to localise and remove, backdoors can persist through safety training. While parameter-efficient fine-tuning methods capture the full parameter changes introduced during training, they are not easily interpretable. We present the full method of TopKLoRA, extending the preliminary proposal of Masiak et al. (2025), an interpretable-by-design LoRA variant in which only the $k$ largest adapter latents are active at each token. Unlike SAEs, TopKLoRA is trained purely on downstream loss, where optimisation pressure incentivises learning functionally causal latents. We demonstrate this sparse LoRA adapter in a sleeper-agent backdoor case study where we create model organisms by fine-tuning gemma-2-2b, Qwen2.5-1.5B, and Qwen2.5-7B on a poisoned instruction-following dataset. For each organism, we identify and ablate a circuit that is both necessary and sufficient to implement the backdoor. Furthermore, our new fine-tuning method produces sparse circuits that are up to an order of magnitude smaller than a dense LoRA baseline, allowing surgical ablation of the malicious capability without sacrificing instruction-following performance. Finally, we propose a technique to construct model organisms with localised circuits using TopKLoRA latents and gradient-routing, which we use to validate the findings of our circuit search algorithm. We consider these findings as a step toward interpretable-by-design architectures that provide faithful units of analysis without compromising model performance. We share our code: https://github.com/interpretable-finetuning/TopKLoRA.

The Turing Test is passed when a human interrogator cannot distinguish machine-generated responses from human ones. Yet behavioral indistinguishability can reflect two fundamentally different failures: a failure of encoding, in which machine and human output distributions overlap so closely that source information is effectively absent; or a failure of readout, in which source information is present but the interrogator fails to extract it. To dissociate these possibilities, we inverted the Turing Test, placing humans and AI systems in symmetrical interrogator roles. Using a quantitative information-processing framework, we separated encoding, the information contained in the text; readout, the extraction of that information by an interrogator; and meta-readout, the interrogator's explicit access to the features guiding its own judgments. Across two series of experiments involving more than 3800 human interrogators and two generations of GPT models (GPT-4 and GPT-5), we found a striking dissociation across processing levels and interrogator types. As AI-generated responses became more human-like from GPT-4 to GPT-5, source information became more difficult for human interrogators to extract. GPT-5 interrogators, by contrast, read out source-discriminating information more accurately than GPT-4 interrogators. This improvement, however, came with a specific limitation: GPT-5 explanations only partially aligned with the feature weightings that actually predicted their judgments. Human interrogators showed the complementary pattern, with modest readout alignment but comparatively higher meta-readout alignment. Together, these findings suggest a double dissociation between readout and meta-readout across human and GPT interrogators.

This paper proposes a five-level organization of message formation and understanding within thinking and activity in which ASI participates. Its central thesis is that a language model's context is fundamentally incapable of presenting the foundations of utterance formation for examination and verification. These foundations are established in advance by acquired schemas, systemic representations of the world of activity, and connections accepted as true within them. Neither context nor trained LLM weights, in themselves, attest to the presence of deliberately constructed systemic representations. The five levels distinguish systemic representations of the world of activity, a particular situation cut out from those representations, its narrative description, the full sense of a message, and text. A situation's schema-model is not identical to a story about it. A recipient reconstructs the full sense from an abbreviated text through jointly constructed and agreed-upon representations and joint activity. Reverse work proceeds from the full sense through an extended description and the situation's schema-model to the reconstruction of systemic foundations. The theoretical foundations draw on the systems methodology of G. P. Shchedrovitskiy, the structural differential of A. Korzybski, and the Meaning-Text theory of I. A. Mel'chuk. The contribution is a specific systemic-activity organization of content for examining and adjusting message foundations and controlling the consequences of their transformation. Further research concerns large graphical models of activity on reconstructible systemic foundations, distinct from text generation from context.

In speech enhancement, standard supervised training requires pairs of noisy and clean speech. Such pairs are typically created by artificially degrading clean speech, leading to a domain mismatch between synthetic training data and real recordings. Noisy target training (NyTT) mitigates this limitation by training an enhancement model to predict a noisy recording from a noisier version generated by adding extra noise. However, since the training target is noisy, NyTT might learn to leave residual noise in the enhanced output. In this paper, we address this issue in NyTT by incorporating a Sinkhorn drifting loss that matches the distribution of the enhanced speech to that of unpaired clean speech. The Sinkhorn drifting loss attracts enhanced speech toward high-density regions of the clean-speech distribution while repelling it from the current enhanced-speech distribution in latent space. This loss uses only unpaired clean speech rather than the clean counterpart of each degraded recording, preserving the weakly-supervised nature of NyTT. Experimental results on VoiceBank-DEMAND and CHiME-3 real recordings demonstrate the efficacy of the proposed combination.

Latent communication requires both an effective continuous reasoning process and an interface through which an external model can leverage its intermediate computation. We present J-Comm, a framework comprising two core components. First, a training-free Geometric Agreement for Latent Reasoning (GEAR) reweights token candidates over existing model embeddings, which monotonically optimizes geometric consensus among candidate tokens in embedding space at each step. Second, we introduce an H+J (Hidden-and-Jacobian) communication interface that gives the Receiver two complementary views of the Sender's reasoning. The H channel supplies continuous inputs derived from feedback distributions, while the J channel conveys intermediate-state cues in vocabulary-indexed Jacobian-lens coordinates. Intuitively, H provides inputs for the Receiver to reason from, while J offers additional guidance within its computation. The Receiver reconstructs these cues using its own dictionary and injects them into intermediate layers during prefill, enabling communication without sharing key-value caches or requiring identical model widths. In Qwen3-8B to Qwen3-4B communication, J-Comm achieves the highest mean score across seven benchmarks, including 94.80% on MATH500, 54.55% on GPQA-Diamond, 73.33% on AIME2024+2025, and 62.93% success on ALFWorld. Compared with TextMAS, it improves performance by 1.29-8.59 percentage points and reduces generated text tokens by 19.2-74.8%.

In LLM-agent opinion simulations, each agent reports the probability that a claim is true after seeing its own previous probability and the probabilities reported by its neighbors, and that number becomes its state in the next round. If the number describes the agent's position rather than the sentence it was shown, then rewriting the claim as its complement and replacing every probability p with 1 - p should return the complementary answer, so the two reports should sum to one. We test this with three small open models: Qwen3-4B, Llama-3.2-3B and Gemma-3-4B. On 2,304 agent states taken from a running network simulation, Qwen's paired reports sum to 1.18 on average, and on 48 real fact-checking claims from AVeriTeC they again exceed one. A controlled benchmark of 24 worlds, each with exactly two possible outcomes, shows the same failure in all three models. The discrepancy is two to ten times larger than the change produced by a same-meaning paraphrase, and it persists under repeated sampling and across peer group sizes and report strengths. When the reports are fed back into two populations that start from complementary states, the populations do not remain mirror images: their divergence is positive after five rounds for every model. Averaging each answer with the complement of its paired answer restores consistency by construction and lowers Brier loss averaged over the two wordings, but at the final round it beats the better of the two original wordings for only one of the three models. A schedule-by-targeting experiment on the same model gives a small, stable interaction because such sweeps keep the claim wording fixed. Consistency under rewording therefore needs its own test before simulated probabilities are read as beliefs

Pipelines that combine large language models (LLMs) with static analysis let the model consume the analyser's output as evidence, but whether the model distinguishes true from false evidence, and whether its scrutiny depends on what the evidence claims, has not been measured. We built an intraprocedural analyser for C/C++ that produces proof obligations and audited vulnerable and patched functions from PrimeVul under six evidence conditions that differ only in the notes, including fabricated assurances that a missing check exists and fabricated alarms that an added check is missing. The study covers 916 pairs in four disjoint sets, eleven model configurations from four vendors and 16,572 audits; no model is trained or fine-tuned. True evidence barely improved accuracy (+0.02). False evidence acted asymmetrically: relative to true notes about the same operations, false assurances lowered the detection of real vulnerabilities by up to 19 points, whereas false alarms never significantly lowered accuracy and were disputed far more often. An internally registered replication on seven new models confirmed that false assurances move verdicts towards their claim more than false alarms do (mean asymmetry +0.16, 95% CI +0.10 to +0.24). Instructions in the audit prompt did not reliably reduce deference; the best one failed to replicate. Verifying each claim in a separate call and showing the refuted claims raised accuracy under false assurances by 12 points in a pre-specified test (p=0.002) without loss under true evidence; for a second model the gain was smaller and not significant. Pipelines should therefore verify a tool's guarantees before the model relies on them, and studies should measure deference in a direction-aware way.

Low-rank SVD-based compression offers a powerful strategy to reduce the computational costs of LLMs. However, existing methods face two key limitations: (i) global rank allocation, where uncalibrated error proxies fail to capture complex error propagation, and (ii) decomposition quality, where Fisher-based estimators suffer from severe rank collapse. In this work, we address these limitations by introducing Layer-wise Error Modeling Search (LEMS) and KFAC-SVD. LEMS advances rank allocation by introducing a layer-wise error surrogate that integrates local and global layer importance alongside a propagation bias, enabling effective global rank allocation via an ILP formulation. KFAC-SVD improves decomposition quality by utilizing token-wise statistics, mitigating the rank deficiency observed in prior Fisher-based SVD approaches. Across Mistral, Qwen3, and Llama 3 model families, we show that LEMS consistently outperforms existing search strategies, delivering significant zero-shot accuracy gains of up to 4.7 p.p. that generalize to model sizes of 70B parameters, while KFAC-SVD achieves an average perplexity improvement of 19%. Project page and code: https://lems-svd.github.io

Echocardiography is the cornerstone of cardiovascular diagnostics, yet its global utility is constrained by a critical shortage of experienced interpreters. Current artificial intelligence (AI) solutions remain limited, typically operating as opaque "black boxes" or focusing on narrow tasks that fail to capture the holistic, guideline-driven reasoning of a clinician. Here we introduce EchoChat, a generative multimodal vision-language model that performs comprehensive echocardiographic analysis to support clinician interpretation. By emulating the cognitive workflow of a cardiologist, EchoChat integrates vast medical knowledge with a chain-of-thought reasoning mechanism to explicitly ground its conclusions in visual evidence. Trained on echocardiography-instruction pairs from 46,358 examinations, our model establishes a new state-of-the-art across diverse cohorts from six international medical centers. It supports the broadest diagnostic scope of 33 cardiac conditions and findings, including 15 newly added conditions for which it achieves an average F1 of 73.7% across available external evaluations. It surpasses existing best-performing foundation models in cardiac condition recognition (improving average recall by 13.54% on shared tasks) and quantification (average R2 score increase of 0.275 on shared tasks). Furthermore, in a source-blinded echocardiographer reader study, echocardiography reports co-generated with EchoChat were rated higher in diagnostic accuracy and completeness by two senior echocardiographers, and improved diagnostic performance from 72.78% to 82.75% in terms of F1 score over routine-reader only reports. These results demonstrate that integrating clinical reasoning into multimodal AI can democratize specialist expertise and transform the standard of cardiovascular care.