Research Papers

Core contribution: Replaced sequential recurrence with multi-head self-attention and positional encodings, enabling parallel training over massive sequence contexts.

Core contribution: Decomposes weight updates Delta W into rank-r products A * B (where r << d), reducing trainable parameters by 10,000x.

Core contribution: To study this challenge, we introduce long-horizon memorization, a setting in which a model learns 100 query-answer tasks through continual supervised fine-tuning without retaining earlier training examples or receiving task identifiers at inference. Sequential updates cause catastrophic forgetting, and no single continual learning mechanism we evaluate maintains strong retention at this horizon.

Core contribution: Beyond playing games, recent systems model players and game dynamics, support design and development, adapt player-facing experiences at runtime, and evaluate resulting artifacts. Yet these directions have evolved largely separately, obscuring which capabilities transfer across settings and which remain tied to particular games, engines, interfaces, or player populations.

Core contribution: We present StepAudio 3 Realtime, an audio-language foundation model organized around a continuous listen-converse-think-act loop. Deep Perception captures rich acoustic cues to interpret user intent, while Seamless Duplex models synchronized audio streams to handle pauses, backchannels, and interruptions naturally.

Core contribution: We first use the Headroom-Closed Index (HCI) to reveal the problems of existing LLMs, then introduce the RSI concept and its development roadmap: from improvement-execution autonomy, improvement-strategy autonomy, experience-acquisition autonomy, and environment-adaptation autonomy, to recursive meta-improvement. Next we examine RSI across scenarios (e.

Core contribution: The StepAudio Music Tokenizer represents audio as a 50-Hz stream from a 65536-entry single codebook, using semantically informed self-supervised and multi-task training to preserve musical structure and reconstruction-relevant information. A flow-matching diffusion Transformer (DiT) predicts continuous StepAudio VAE latents, which our VAE decoder converts into 48-kHz audio.

Core contribution: We introduce RSIAgent, a training-free multi-agent framework for recursive self-improvement through autonomous memory construction. RSIAgent coordinates curriculum, actor, and verifier agents to continually explore the environment, validate outcomes, and retain environment-specific knowledge, including reusable causal relationships between actions, conditions, and consequences.

Core contribution: Grouped-query attention (GQA) reduces this cost by sharing key-value heads, but still stores both a key and a value at every step. We introduce Grouped Value Attention (GVA), which stores grouped values and reconstructs content keys with a learned linear map.

Core contribution: Agentic visual creation provides explicit references, editable 3D scenes, or executable game states for stable control, but does not by itself guarantee high object or character fidelity. Combining the two can enable stable, high-quality generation.

Core contribution: Existing approaches face a representation trade-off: perspective models operate on local views and must preserve off-screen content over long rollouts, whereas broader spatial coverage is typically obtained by synthesizing full-sphere videos or constructing explicit 3D representations. Motivated by the complementary roles of global context and selective local acuity in visual perception, we present AlayaVista, a camera-controllable streaming video world model that decouples panoramic world evolution from perspective observation synthesis.

Core contribution: We introduce Agent as Policy (AGP), which places task planning and execution under the agent's control. Given a task and a robot interface, the agent interprets visual evidence, writes executable programs, issues motion commands, and revises its actions in response to physical outcomes.

Core contribution: RL shows large improvements on easy problems that an LLM is already good at solving, but small improvements on hard problems. We call this the Matthew Effect in RL for LLMs, after the phenomenon of cumulative advantage from economics and network science summarized as "the rich get richer".

Core contribution: Off-policy methods such as supervised fine-tuning, on the other hand, can leverage external knowledge beyond the base model's capabilities, but may suffer from large distribution shift. The key challenge is thus to expand exploration without sacrificing learnability.

Core contribution: We bring these predictions into action generation and selection through a shared trajectory model. Dynin-Robotics implements this formulation on Dynin-Omni, an omnimodal masked-diffusion backbone, representing language, visual observations, goals, and actions as discrete tokens.

Core contribution: With the increasing tendency of incorporating technology into environments and activities in everyday life, new methods are being proposed to better integrate devices and mankind.Networks designed for monitoring areas through video and image systems are being implemented in several applications.Recent studies have shown interest from both the academy and industry to integrate wireless surveillance networks with lowcost and long range transmission technologies, such as LoRa (Long Range).With the intention to explore this topic, this work presents the development of a system prototype for intelligent monitoring as a basis to future implementation of low-cost LoRabased Wireless Image Sensor Networks (WISN).The developed system consists of a processing unit, a data routing interface and multiple sensor nodes.Its operation is autonomous and initiated through the detection of human presence, by the sensor node, capturing an image of the individuals present at each detection.The data is transmitted through LoRa devices to the central hub, where python-based methods of facial detection and recognition are employed.Several tests were performed to both adjust the system and to verify its efficiency.The results achieved indicate the viability of the proposed low-cost LoRa-based WISN.

Core contribution: Existing hallucination evaluations are mostly claim-centric; they assess whether generated statements are supported, but not whether evidence remains traceable through rationale, confidence, and final action. We introduce E2A-Bench, a 969-query benchmark for financial chart reasoning, constructed from 323 HS300 constituents under three input modalities with deterministic OHLCV-derived evidence anchors.

Core contribution: Rather than treating this drift as an uncontrolled consequence of optimization, we specify a behavioral drift budget before optimization and ask how to boost the target-task performance within it. Locally, behavioral drift induces a shared geometry anchored at the reference model, with the drift budget defining a boundary within this space.

Core contribution: However, when these specialists are trained solely on question--answer pairs without explicit reasoning supervision, what governs the trajectories they generate? In this work, we show that specialist optimization implicitly selects from this latent trajectory space.

Core contribution: Do reasoning models robustly exhibit such systematicity? If so, we would expect consistent performance on structurally equivalent variants of the same task.

Core contribution: However, most existing learning algorithms focus on predictive performance, overlooking the joint optimization of other desirable properties, such as structural sparsity. In this work we propose TREVIS, an approach for learning decision trees with respect to complex objectives, based on the exploration of the latent space of a Tree Transformer Variational Auto-Encoder (TTVAE).

Core contribution: To address this gap, we present UltraInstinctVR, a novel tool that automates the generation and execution of concrete system-level tests for VR applications, helping developers efficiently validate interactions and behaviors.

Core contribution: Urban trees face increasing extreme heat, yet the timing of their short-term physiological responses remains unclear.

Core contribution: This paper critically reflects on the energy costs of using genAI in design education, using a workshop held in 2023 with 49 students as a springboard for reflection.

Core contribution: We present DoubleAgents, a system for human–agent alignment in coordination tasks, grounded in distributed cognition.

Core contribution: The French Open Science Strategy (SSO) 2026–2033 aims to strengthen the transformations initiated with the two previous national plans.

Core contribution: In this work, we present a multirobot aerial manipulator for floating litter collection, leveraging the capabilities of aerial robots.

Core contribution: In this work, we present a scalable and multi-purpose pressure sensing system for multidirectional monitoring of tissue interactions, establishing a robust solution for deploying diagnostic and therapeutic instruments in various types of endoluminal interventions.