As AI agents become increasingly capable of generating and optimizing hardware circuits, ensuring the correctness of AI-generated designs becomes more important. Meanwhile, advances in AI lower the barriers to developing and applying formal methods, creating new opportunities to automate and improve hardware verification. This invited paper presents an overview of the methodologies and applications of hardware formal methods in the AI era. First, we examine how agentic AI advances hardware formal methods across three levels: specification, proof construction, and reasoning engines. Second, we discuss how formal methods support reliable AI-driven hardware development through RTL generation, debugging, and correctness-preserving optimization. We review representative research alongside our recent contributions, highlighting key technical advances and insights from experimental results. Finally, we discuss future research opportunities toward proof-native hardware development, where formal reasoning could become an integral part of autonomous hardware design.
Existing sequential user modeling methods mainly focus on improving the backbone encoder or learning task-specific user representations for a single prediction objective. We instead study how to export the multi-layer user states of a frozen long-history backbone as compact, timestamp-addressable representations for downstream training backfill and serving. In this setting, the user state at a given timestamp summarizes the user's actions before that time, so downstream systems need only the state aligned to each target timestamp rather than the full hidden-state timeline. To this end, we present \texttt{SeqFold}, a compressed-state interface that combines learned multi-channel fusion, channel-specific temporal decay, and Lookup-Free Quantization (LFQ) to produce compact discrete codes from frozen backbone states. Task-specific decoders then map these shared codes into flexible embedding spaces for downstream models. To make timestamp-addressable export practical at scale, we further develop a fused streaming implementation that updates the fusion recurrence after every action while materializing states only at target timestamps. On large-scale proprietary datasets, SeqFold reduces per-state representation storage by $3{,}400\times$ while preserving most of the downstream ranking performance of the uncompressed model, showing that SeqFold is an effective interface for reusing long-history sequence backbones in downstream models.
We investigate how the persistent topology of neuron- weight point clouds evolves under stable and non-stationary supervision. An identical multilayer perceptron is trained on Fashion-MNIST with correct labels, a fixed random permutation of the training-label vector, or a fresh permutation every epoch. Five paired seeds share initialisation and minibatch order, yielding 15 models trained for 100 epochs. Raw and row-normalised incoming weight clouds are analysed with H0 and H1 Vietoris Rips persistence at 22 checkpoints. Correct-label models reach 89.39% mean true-label test accuracy; fixed-random models fit 67.94% of their assigned training targets without generalising to the original task. Correct-label models have descriptively greater time-averaged normalised fc3 H1 persistence than either random-label condition, but the configured exact comparisons do not reach conventional significance after Holm correction. Early condition classification achieves 60.0% mean balanced accuracy using topology, 66.7% using conventional diagnostics, and 60.0% using both. Thus, no incremental predictive improvement is demonstrated in this setting. The archive contains all 3,300 configured topology cases. Interpretation is limited by five seeds, a single architecture and dataset, 282 capped H1 distance comparisons, and deferred bottleneck analysis.
Multi-camera systems support performance recording, sports and human-movement analysis, where selecting an informative viewpoint is important for reliable human pose estimation. We propose the Candidate-Joint Reprojection Ranking Network (CJRR-Net) for future-view selection among eight calibrated cameras in MPI-INF-3DHP. The model combines current-view pose observations, camera configuration, and candidate-specific geometric features obtained by reprojecting the estimated 3D pose and its uncertainty into each candidate view. CJRR-Net achieves a selected-view 2D mean per-joint position error (MPJPE) of 105.18 pixels, compared with 114.64 pixels for the shared-input MLP and 117.56 pixels for adapted PoseErrNet. It also achieves the lowest mean error among the listed policies in each evaluated occlusion group. A blender rollout illustrates repeated camera selection. These findings support candidate-specific geometry for single-view selection in the evaluated fixed eight-camera setting.
Multi-agent systems provide a convenient framework for the massive deployment of agents based on Large Language Models (LLMs). It provides a sound framework where agents can interact with existing software to perform actions in an autonomous way. Agentic AI is the new term to denote these AI-based autonomous systems. In real multi-agent systems, agents need to deal with malicious agents some of which can exploit security vulnerabilities to do harm. A memory poisoning attack is an example of exploiting a vulnerability to modify agents knowledge and, thus, affect its behavior. In an agent, there are different types of memory systems (semantic, episodic, and short-term) which have their own characteristics and for which different memory poisoning attacks may be developed. In this paper we focus on semantic memory, we discuss its attack, and mitigation strategies. We introduce private knowledge retrieval and private inference. We propose a signature system in the knowledge base to prevent from undesired manipulation. This should avoid memory poisoning attacks by introducing security by design in the construction of the semantic memory. In our architecture, an agent contains a local Prolog-like inference system (i.e., using Horn clauses to represent knowledge) which accesses an external knowledge base to get the required knowledge. This externally signed knowledge base is accessed privately. We provide different implementations in Python using different privacy levels including k-anonymity and information-theoretical privacy (IT-PIR).
Rapid AI data-center power fluctuations challenge grid integration. Sustained grid-import smoothing requires allocating limited battery power and energy under uncertain demand. We formulate risk-aware BESS scheduling that coordinates advance reserve commitments and demand-responsive dispatch, accounting for operating costs, grid-ramp penalties, and their conditional value-at-risk. We develop battery scheduling decision-focused learning (BS-DFL) with a constraint-aware neural optimization layer trained directly from this objective without optimal-solution labels. The trained layer is then frozen to guide demand-predictor learning through scheduling costs and ramping risk, enabling solver-free rolling-horizon scheduling. In a 10,000-minute simulation using loads derived from measured AI workloads, BS-DFL reduces mean absolute minute-to-minute grid-import changes by 71.70% and total scheduling cost by 18.01% relative to an MSE-trained predictor coupled with a mixed-integer linear programming solver, with an average decision time of 18.77 ms.
We present Walrus, a unified 8B speech-and-audio foundation model handling speech recognition (including code-switching and multi-talker speech), speech-to-text and speech-to-speech translation, text-to-speech, emotion recognition, audio captioning, and general audio understanding. Our main contribution is not a new architecture but a data-centric training recipe; we use a standard decoder-only backbone initialized from Gemma 4, so that gains can be traced to the data. For speech generation, Walrus predicts low-bitrate discrete speech tokens, and a lightweight token-to-waveform decoder converts them into high-fidelity audio. The recipe is a three-stage data curriculum that moves from broad multilingual data to multi-talker, code-switching, and domain-specific data. At each stage, task-aware and dataset-aware weighting balances tasks with very different token budgets and protects core skills from forgetting. Trained on over 600M samples covering more than 20 tasks and 100 languages, Walrus leads on most multilingual, multi-talker, and code-switching ASR, speech translation, and audio understanding benchmarks, outperforming both open speech language models of comparable size and dedicated task-specific models. On most benchmarks, it even matches or surpasses the $4\times$ larger Qwen3-Omni's performance. Results suggest that data engineering is a primary driver of quality in multi-task speech and audio modeling.
Answering spatial questions about multi-source scenes requires a model both to identify each sound and to locate it. Existing spatial audio-language models obtain spatial information from a dedicated encoder, which separates spatial and semantic processing into distinct feature spaces. We study whether a general-audio codec trained without spatial supervision can serve both purposes. A controlled layer-wise analysis shows that the frozen codec retains spatial information, with lateral cues carried by the difference between the two ear channels. This information, however, is not readily accessible: in dual-source scenes, a model trained on these features predicts left/right at chance and, on the remaining axes, assigns the detected direction to the queried and the non-queried source equally often. We attribute this failure to an audio--text information asymmetry, in which the audio contains both sources while the answer describes only the queried one. A controlled $2\times2$ experiment shows that either completing the scene description or exposing the inter-channel difference lifts left/right accuracy above chance, whereas binding directions to the queried source requires both. Building on these findings, AURA keeps the codec and the language model frozen and exceeds the strongest dedicated system on SpatialSoundQA by 10.31 points in dual-source direction accuracy, with about 40\% of its trainable parameters.
Media bias understanding and detection remains an under-researched problem for Indian context, specifically for Indic languages, such as, Hindi, which is widely spoken in India. In this paper, we introduce a comprehensive, multi-annotator, explainable dataset consisting of 448 unique articles from four leading Indian news agencies publishing in Hindi. We focus on Hindi, the most spoken and an official language of India. The study of media bias is particularly relevant in the Indian context due to the country's linguistic diversity and large population, where the same event is often reported by multiple news agencies in different languages, thereby introducing scope for bias. The dataset comprises Hindi political news articles annotated by 14 experts and classified into four bias categories - selection bias, framing bias, clickbait bias and omission bias - when an article is identified as biased or unbiased. For each biased article, annotators also provide a textual explanation, enabling future semi-automated approaches to incorporate explainability when language models reason over the dataset. Cohen's Kappa scores were 0.81 for simple bias/no-bias classification reflecting strong inter-annotator agreement. For the multi-label task, the Cohen's Kappa values are 0.89, 0.75, 0.81, and 0.79 for clickbait, selection, framing, and omission biases, respectively, resulting in an average Kappa of 0.80, which also reflects strong agreement.
In this paper, we investigate how human egocentric videos can benefit robot policy learning without action supervision. Building on a pretrained VLM, we propose Shared Prediction on Action and Representation (SPAR), an MoT architecture in which future visual state prediction can provide independent and complementary learning signals alongside action prediction. We find that representations learned from egocentric videos improve downstream performance on out-of-distribution robot tasks, with further gains as video data scales. We also study the effects of objective choices and data curation, providing insights into how to leverage action-free videos for scalable robot learning.
We introduce the Functional Preservation Score (FPS), a null-normalised, behaviourally grounded measure of how much of a neural network's input-output function survives a structural operation, and use it to audit the operations underlying speculative "mind-migration" pipelines on public data that runs on a single laptop. Across five pre-registered studies on the C. elegans connectome, whole-brain calcium activity, and hippocampal replay, FPS yields two positive characterisations and a three-part negative. Low-bit quantisation of a real connectome exhibits a structure-dependent sparsity floor: uniform int4/ternary quantisation is forced to roughly 73%/98% effective sparsity because heavy-tailed integer synaptic weights round to zero, while mixed-precision protection of hub neurons lowers these floors to about 27%/32%. Decoded hippocampal replay sequences compress heavily yet remain identifiable at a cross-validated top-1 accuracy of 0.69 (95% CI [0.63, 0.74]; chance 0.125; permutation p < 1e-3) from latents of order 10-50 bits per event, and a bounded incremental store shows a graceful-versus-cliff capacity trade-off. The migration pipeline fails in simulation at each stage for largely independent reasons: the putative "identity payload" is not a functionally separable module (p >= 0.16 at every tested size); cross-individual functional transport via Gromov-Wasserstein collapses once the target's wiring is respected (fidelity 1.00 to 0.40, [0.31, 0.49], p < 1e-3) or alignment is imperfect (0.06, [0.00, 0.13]); and biologically plausible local plasticity cannot reach the function the target topology provably admits (gap 0.45, [0.27, 0.64], p < 1e-3, d = 2.0). We release FPS, the data loaders, and all operations as a toolkit. All results are at nematode / small-network scale and concern coherence in simulation, not the feasibility of reading or writing a living brain.
On-policy distillation (OPD) improves language models by distilling teacher predictions on student-generated trajectories, while on-policy self-distillation (OPSD) removes the need for an external teacher by using a reference-conditioned copy of the model as the teacher. Despite its great early improvements, we find that OPSD does not remain consistently effective under continued training and can eventually suffer substantial performance degradation. We study this training behavior and link the degradation to a growing mismatch between the evolving reference-free student and the frozen privileged teacher. In the early stage of OPSD, the privileged teacher remains closely aligned with the student and provides useful guidance for improving the student's generation trajectories. However, as training progresses, the growing distributional mismatch between the privileged teacher and reference-free student can cause the teacher to provide misleading supervision, ultimately degrading student performance. This phenomenon motivates us to first leverage privileged reference supervision to acquire useful knowledge and then transfer it to a fresh student. We introduce Bootstrapped On-Policy Distillation (BOPD), a two-stage framework that first uses OPSD to internalize task-relevant information from reference-conditioned supervision into an intermediate model, and then distills this reference-free intermediate model into a fresh copy of the initial model. Experiments across models show that BOPD improves mathematical reasoning and code generation beyond both the initial models and the constructed OPSD teachers. These results suggest that privileged reference supervision can be more effectively exploited by first transferring its task-relevant signal into a task-adapted model and then distilling from that model without reference conditioning.
Retrieval-augmented generation (RAG) is bottlenecked by the token budget of large language models: feeding long documents into the context window is expensive, while aggressive textual compression discards fine-grained information that is irreversibly lost in a one-dimensional token sequence. We argue that this bottleneck is not fundamental to information density but to the choice of compression \emph{modality}. Modern visual encoders can pack the contents of a rendered text page into roughly a hundred visual tokens with little semantic loss, opening a new compression-fidelity tradeoff that purely textual methods cannot reach. We present \textbf{OpticalRAG}, a framework that renders documents as page images, encodes them with a frozen vision encoder, and distills the retrieved visual tokens into as few as $16$ injected tokens before passing them to the LLM. Two ingredients make this work without expensive cross-modal training: (i) \emph{Encoder-Bridged Retrieval}, which projects text-query states into the visual feature space of the frozen encoder, through a projection shared with the compressor, to retrieve visual tokens from a text query, with no contrastive training; and (ii) \emph{Query-Driven Distillation}, a lightweight cross-attention compressor that condenses the retrieved visual tokens into a small set of query-conditioned prefix tokens. On LongBench-v2, BAMBOO, and LooGLE-v2, OpticalRAG achieves the best compression-aware accuracy (CAP) on all three benchmarks while injecting only $16$ tokens, and improves the closed-book Qwen2.5-7B-Instruct backbone by $5.17$ points on LongBench-v2.Code is available at https://github.com/brighterr/OpticalRAG.
Reasoning about people's plans and goals based on their actions is fundamental to human social interaction and essential for cooperative, embodied artificial intelligence. Formalizing this capacity in naturalistic, embodied settings has been a longstanding challenge. Here, we present GenEA, a novel agent representation capable of reasoning over multi-step, long-horizon embodied plans. At the heart of GenEA lies a multilevel state-space that couples the coarse biomechanics of an articulated body and its interactions with nearby objects by extending the task-and-motion planning framework from robotics with simulation-based intuitive physics. By embedding this state-space in a generative model, GenEA yields a joint probability distribution over goals and embodied plans. Across several experiments spanning the qualitatively distinct domains of object manipulation and navigation, featuring diverse tasks and multiple behavioral measures, we show that GenEA, through simple transformations of this joint distribution, recapitulates how humans reason about embodied plans. Crucially, GenEA consistently outperforms strong alternative models, including plausible heuristics, standard task-and-motion planning, and a state-of-the-art vision-language model. GenEA thereby offers a new perspective on the architecture of human reasoning and social cognition, suggesting that embodied artificial intelligence systems seeking to seamlessly interact with humans should adopt GenEA-like generative models of action planning.
Selecting an LLM for next-step recommendation requires benchmarks that reflect the intended educational task and domain. We present a benchmarking method that uses Knowledge Space Theory (KST) as a formal evaluation reference. Given a task catalogue and an existing prerequisite relation, the method derives admissible knowledge states and the complete set of immediate next steps for each state. Recommendations generated by a model under a specified prompt are compared with these reference sets. A sample evaluation study uses 2,400 responses from six models across four mathematics domains, four prompt conditions, and two catalogue orders. Exact-set accuracy was 61.3% with descriptions only, 62.8% when models were asked to reconstruct prerequisites, 96.3% with supplied prerequisite relations, and 98.2% with the complete knowledge space. Performance varied across models and domains, and recommendation sets agreed across the two presentations in 85.25% of pairs. The benchmark supports comparisons of model-prompt combinations against a context-specific reference. Its scores describe structural recommendation accuracy under the evaluated conditions; they do not establish learning benefit or the necessity of using an LLM for selection.
Sepsis is a leading cause of ICU death with heterogeneous onset and course, yet most warning models predict only whether sepsis will occur, leaving clinicians uncertain of its onset context and early trajectory once an alarm sounds. To answer these questions on one hourly clock from ICU admission, we built SepsisFlow: before onset, it combines a Mamba model pretrained on 2.57 million EHRs with about 1,700 hourly features in gradient-boosted trees to predict 6/12/24-h risk; at onset, it fuses Mamba and tabular encoders via a gated network into continuous patient states defining onset contexts; after onset, it encodes 72-h organ-support and physiological trajectories via an attention bidirectional recurrent network into early courses. For pre-onset warning in MIMIC-IV, SepsisFlow detected more cases than NEWS2, qSOFA, and SIRS at equal alarms, gaining about 8 cases per 100 from hourly updating, while outperforming 16 retrained algorithms and maintaining risk ranking in eICU. At onset, its contexts separated 2% cardiovascular-ICU from 15%-17% other 28-day mortality, and within each context after onset, persistent support carried fivefold higher mortality than de-escalation. SepsisFlow establishes a foundation-model paradigm for continuous risk and trajectory stratification in critical care.
The complex interactive dependencies between human behavior and 3D scenes make accurately predicting human intent a major challenge for human motion prediction (HMP). While recent studies use human gaze to infer intent, gaze signals suffer from ambiguity, drift, and a reliance on eye-tracking hardware. Research in cognitive science indicates that egocentric views inherently encode potential interactions, while head motion serves as a crucial precursor in motion planning. Motivated by this, we propose EgoHMP, a novel framework that uses egocentric views and head motions as robust carriers of interaction intent. By translating this intent into an envisioning of future interactions, EgoHMP achieves precise HMP in 3D scenes. EgoHMP first employs a vision-motion alignment encoder to align the visual and motion feature spaces. Based on these aligned features, an egocentric intent predictor with a modality-aware modulation mechanism adaptively suppresses invalid interaction cues to predict an accurate interaction probability map. Furthermore, to enhance the stability and realism of the generated motions, we propose a contact-aware motion decoder. Specifically, inspired by the human cognitive process of planning trajectories prior to action execution, we first predict a global trajectory to serve as a self-prompt, effectively mitigating the optimization conflict in diffusion models. Subsequently, we introduce a multi-stage denoiser, which relieves motion artifacts by incorporating human contact modeling through a multi-stage optimization strategy. Finally, a GCN-based motion decoder is employed to synthesize physically plausible and semantically consistent motions. Extensive experiments demonstrate that EgoHMP achieves state-of-the-art performance.
Drug discovery involves multiple interconnected research stages, including molecular design, property prediction, and virtual screening. These processes are characterized by complex workflows, interdependent computational tasks, and diverse specialized tools, placing substantial demands on the coordination of scientific models and computational resources. However, existing large language model (LLM)-based scientific agents still face challenges in autonomous planning of complex tasks, adaptive selection of specialized models, and reliable execution of multi-stage scientific computations, limiting their effectiveness in automating complex drug discovery workflows. To address these challenges, we propose DrugCoding, a scientific programming agent system that integrates scientific task planning, adaptive model routing, multi-agent collaboration, and programmatic tool execution. DrugCoding enables autonomous task planning and dynamic model adaptation for complex research objectives. It can flexibly invoke specialized scientific models, autonomously generate and execute code according to task requirements, and leverage multi-agent collaboration to support efficient execution and reliable validation of complex scientific tasks. To evaluate DrugCoding's performance in scientific tool invocation and multi-stage task execution, we conduct multi-turn tool-calling experiments across diverse language models and scientific tool interfaces. Experimental results demonstrate that DrugCoding achieves an average session-level Pass@1 of 97.35% on the full benchmark, outperforming the strongest baseline, Codex, by 9.24%. On the challenging subset, DrugCoding achieves 97.92%, exceeding the strongest baseline.
Predicting gene expression directly from hematoxylin and eosin (H&E) images offers a scalable and cost-effective approach to deriving spatial molecular insights from routine histology. Pathology foundation models, typically pretrained using tile-level objectives, may not adequately capture the fine-grained cellular features required for single-cell molecular prediction. In this work, we systematically evaluate and establish domain-informed and task-specific strategies for adapting pathology foundation models to single-cell spatial gene expression prediction. Using paired H&E and Xenium 5K data, our approach improves mean gene-wise Pearson correlation by 2.67x over the standard zero-shot baseline and achieves state-of-the-art performance on this task. We demonstrate strong gene-detection performance across the full gene panel (AUROC = 0.800) and importantly identify that predictive performance varies substantially across biological contexts, with our model achieving its highest Pearson correlation for spatially variable genes (r = 0.643). Our findings demonstrate that, despite the challenges of sparse single-cell measurements, H&E-derived representations can recover biologically meaningful spatial expression patterns, highlighting both the opportunities and limitations of morphology-based molecular inference at cellular resolution.
Transformer feed-forward blocks combine several design choices: squared activations, gating, and tied versus untied parameter sharing, whose separate contributions to expressivity remain poorly understood. We introduce CP-MLPs, a tensor framework in which each multiplicative unit contributes a rank-1 interaction with three roles: detecting an input feature, reading a value, and writing to the output. Tied gated blocks such as ReLU$^2$ force the detector and value roles to share the same direction; untied gated blocks such as ReGLU and SwiGLU let them differ. We show that this distinction yields exact, degree-wise tensor-rank separations: untied detector-value units are strictly more width-efficient for indefinite quadratic interactions and role-asymmetric monomials, while tied units are structurally matched to symmetric pure-power targets. The width advantage of untying therefore depends on the target structure. Experiments show loss plateaus below the exact-representation width, but these plateaus alone do not establish a positive approximation gap: arbitrarily accurate approximation is governed by border rank. In trained gated MLPs, untied blocks retain an advantage in selected comparisons at matched input-projection budget. Code is available at https://github.com/rarefin/CP-MLPs.