arXiv subjects

Mauro Martino

Publications and source records attributed to Mauro Martino.

17 recordsLinked to original sources

arXiv2026

Post-Training Large Language Models with LLMON

Large Language Models (LLMs) treat all context tokens uniformly, lacking explicit separation between executable instructions and passive data. This leads to prompt injection vulnerabilities, ambiguous instruction binding, and unreliable execution control. We present a post-training methodology using LLMON (LLM Object Notation), a framework that enables separating instructions from data and requires explicit identifier-based execution. We construct a large-scale structured training corpus of 4.6B tokens spanning general, RAG (Retrieval-Augmented Generation), and tool-calling data, augmented with distractor-infused variants in which competing instructions and irrelevant data segments are present in context but not referenced. Post-training two Granite-4.1 base models (3B and 8B) through supervised fine-tuning followed by reinforcement learning, we show that LLMON-structured supervision reduces attack success rates on indirect prompt-injection benchmarks from 53-100% to 0.02-5.6% while preserving general capabilities. We further introduce attention-based interpretability metrics that diagnose failure modes at individual transformer layers for structured inputs, revealing how our approach reshapes internal attention patterns. Our results demonstrate that explicit execution semantics can be taught through standard post-training without architectural modification.

cs.AI↗

arXiv2026

Flow Reasoning Models: Turning Flows Into Efficient Recurrent Reasoners

Structured reasoning requires making and revising interdependent decisions to reach a globally consistent solution. Existing architectures struggle with this: autoregressive models commit sequentially and cannot revise earlier decisions, while masked diffusion models often require careful decoding schemes to coordinate interdependent predictions. We introduce Flow Reasoning Models (FRMs), a novel framework for structured reasoning that adapts continuous flows over discrete structured outputs with a simple recurrent refinement mechanism. By self-conditioning a flow model on its own past outputs, we turn one-shot denoising into iterative solution refinement. This lets FRMs make and revise decisions in parallel, efficiently coordinating interdependent choices across solutions. Yet conventional self-conditioning becomes unreliable at greater recurrent depth due to exposure bias between one-step training predictions and recursively generated inference states. We address this mismatch with Fixed-Point Forcing (FPF), which trains FRMs on states produced by their own inference dynamics while preserving the standard flow-matching objective. FRMs achieve solve rates of $99.5\%$, $100.0\%$, and $99.9\%$ on Sudoku-Extreme, Zebra, and Maze-Unique, respectively. On Sudoku-Extreme, FRMs achieve higher peak accuracy than the evaluated masked-diffusion and specialized reasoning baselines while remaining highly compute-efficient, matching the next-best method's $98.7\%$ peak solve rate with $44\times$ fewer inference FLOPs.

cs.AI↗

arXiv2026

POCI-Diff: 3D-Layout Guided Diffusion for Controllable Synthetic Surveillance Data Generation

Training robust visual surveillance models requires large-scale datasets with precise spatial annotations, yet collecting real surveillance data is costly, privacy-sensitive, and often legally constrained. Synthetic data generation offers a compelling alternative, but existing methods lack fine-grained 3D control over object placement and appearance, limiting geometric consistency across camera viewpoints. We introduce POCI-Diff (Positioning Objects Consistently and Interactively), a framework that generates annotated synthetic scenes from explicit 3D bounding-box layouts with per-object semantic control. By integrating Blended Latent Diffusion with depth-conditioned ControlNet, POCI-Diff synthesises complex multi-object scenes in a single forward pass, binding individual text descriptions to specific 3D locations. We further propose a warping-free editing pipeline supporting object insertion, removal, and transformation via regeneration, enabling efficient scene variation for data augmentation. Object identity across edits is preserved by conditioning on reference images via IP-Adapter, ensuring appearance consistency throughout interactive scene manipulation. Experiments show that POCI-Diff outperforms state-of-the-art 3D layout-guided generation methods in visual fidelity and layout adherence, while eliminating warping-induced geometric artifacts.

cs.CV↗

arXiv2026

Billions of Sketches Reveal Hidden Cultural Variation in Human Concepts

Claims about the universality of human concepts have been predominantly assessed through linguistic similarity across languages and cultures. However, words are effective as communication devices because they compress rich experiential variation into shared conventions, potentially obscuring hidden individual and cultural differences in how concepts are mentally represented. Here, we analyse 2.6 billion human-made sketches of common concepts from 236 countries and territories to examine conceptual structure through people's visual imagination. Consistent with recent work on image-based cognition, we find that single concepts unfold into multiple distinct visual exemplars, revealing latent information about similarities and differences in conceptual structure across cultures. This variation is strongest for concepts involving haptic interaction, suggesting that visual imagery reflects variation in embodied experience as much as conventional definitions. Comparing embedding models of sketches with word embedding models across languages, we find that their geometries diverge, with visual representations preserving rich semantic and cultural structure that language models compress. Cross-cultural similarities derived from sketches align 32% more closely with established cultural distances than do text-based measures. Together, these results suggest that patterns of human conceptual universality may depend critically on the modality through which concepts are measured, with large-scale sketching providing a direct, high-resolution probe of conceptual diversity across embodied and cultural dimensions of thought.

cs.CY↗

arXiv2026

Hurwitz Quaternion Multiplicative Quantization for KV Cache Compression

We propose \textbf{Hurwitz Quaternion Multiplicative Quantization (HQMQ)}, a \textbf{calibration-free} method for KV cache compression of large language models. HQMQ treats each 4-element chunk of K or V as a quaternion and quantizes its unit direction to the \emph{product} $q_p \cdot q_s$, where $q_p$ ranges over the 24-element Hurwitz group $2T$ (the 24 vertices of the 24-cell on $S^3$, pairwise angle $60^\circ$) and $q_s$ ranges over a per-(layer, head) secondary codebook of $S$ \emph{random} unit quaternions. The multiplicative composition yields $24S$ effective codewords at $S$ stored parameters; random initialization suffices because left-multiplication is an $S^3$ isometry, so seeded codebooks vary in end-task ppl by $<1.5\%$. A per-batch median-multiplier outlier extraction step ($C{=}3$, no calibration) handles modern outlier-heavy architectures. We evaluate on five modern open models: Mistral-7B (dense MHA), Llama-3-8B and Qwen2.5-7B and Qwen3-8B (dense GQA), and gpt-oss-20b (sparse MoE). On Mistral-7B and Qwen3-8B, HQMQ matches fp16 within $0.02$--$0.03$ ppl points at $\sim$5 bits. On Qwen2.5-7B and Qwen3-8B, where naive int4 collapses to $10^4{+}$ ppl, HQMQ + Med3$\times$ recovers fp16 quality within $0.02$--$0.10$ ppl points at $\sim$5 bits. HQMQ Pareto-dominates naive int by $3$--$1900\times$ at matched bits across all five models, and downstream zero-shot accuracy matches fp16 at $3.79$ bits on Mistral. Against the strongest calibrated KV-quantization baseline, HQMQ at $3.79$ bits matches KIVI-4 ($\sim 4.5$ bits) within ${\sim}1$ pt on CoQA, $0.6$ pts on TruthfulQA, and $2.3$ pts on GSM8K, at $16\%$ fewer bits and without a calibration pass. At the storage level, HQMQ delivers up to $5.05\times$ KV compression, shrinking a Llama-3-70B 128k-context cache from 43 GB to 8.5 GB.

cs.LG↗

arXiv2026

Tensor Memory: Fixed-Size Recurrent State for Long-Horizon Transformers

Transformers process images and videos by flattening space and time into long token sequences. While attention and KV caching preserve past features, their memory grows with sequence length and they lack an explicit, persistent spatial state, making long-horizon video understanding and occlusion-sensitive reasoning difficult. We propose Tensor Memory, a lightweight module that augments Transformer blocks with a fixed-size recurrent 3D memory tensor: tokens write into a voxel grid via a differentiable soft write that deposits content as a Gaussian-weighted volume around a predicted continuous 3D location, the memory is updated with an efficient local interaction operator and gated recurrent dynamics, and tokens read back context via continuous sampling with gated residual fusion. Because the memory tensor has a constant size, Tensor Memory decouples state capacity from input length while preserving a spatial inductive bias. We evaluate the module on standard language, image, and video benchmarks and on a controlled toy diagnostic suite designed to isolate when persistent state is beneficial; it integrates with standard Transformer training pipelines and can be attached to or removed from existing blocks without other architectural changes.

cs.CV↗

arXiv2026

Tensor Cache: Eviction-conditioned Associative Memory for Transformers

Autoregressive Transformer KV caches grow linearly with context length; sliding-window caching bounds memory but discards evicted tokens entirely, so relevant evidence outside the window becomes inaccessible. We introduce \emph{Tensor Cache}, a two-level cache that pairs sliding-window softmax attention as a first-level cache (L1) with a fixed-size outer-product fast-weight memory as a second-level cache (L2) fed by KV pairs evicted from the window. Recent tokens remain in exact local attention; evicted pairs are compressed into a per-layer matrix $A$ and read by future queries through a single matrix multiplication, exploiting the linear-attention identity $q_t(k_i \otimes v_i)=\langle q_t,k_i\rangle v_i$. A learned scalar gate fuses the L1 and L2 outputs, and per-head decay and write-rate parameters are trained end-to-end. The outer-product memory and the read identity are well-known; our contribution is their use as an L2 cache fed exclusively by sliding-window evictions, plus identifying that the common chunked-mean training shortcut $A\!\leftarrow\!λA\!+\!η(\bar k\!\otimes\!\bar v)$ silently introduces $C^2{-}C$ spurious cross-token outer products per chunk, and closing the gap with a parallel weighted-sum scan equivalent to per-token writes within float32 epsilon. Across systems scaling, controlled associative recall, long-context language modeling, and memory-capacity diagnostics, Tensor Cache improves the memory--quality frontier over bounded-state baselines.

cs.LG↗

arXiv2026

Small Models, Smarter Learning: The Power of Joint Task Training

Multi-task learning improves generalization, but when does it reduce the model capacity required to learn? We provide a systematic study of how joint training affects the learning transition, the minimum model size at which a task can be learned, using nested arithmetic (ListOps) and permutation groups as controlled testbeds. Certain task pairings dramatically reduce model size requirements: combining easy operations (MAX, MIN, PROD) with hard ones (modular addition, permutation products) enables learning with 2-7 times fewer parameters. Crucially, we also identify when synergies fail: pairing structurally similar hard tasks (e.g., ADD with alternating-sign NADD) provides no benefit, nor does pairing tasks lacking shared computational primitives. PCA of learned embeddings reveals that successful joint training induces structured number representations (ordering, parity, modular structure) absent in single-task models. Transfer experiments confirm these representations are causal: models pretrained on easy tasks learn addition at 7 times smaller sizes. Our results establish that task compatibility, not mere diversity, determines whether joint training reduces capacity requirements, providing quantitative guidance for curriculum design.

cs.LG↗

arXiv2026

ESPLoRA: Enhanced Spatial Precision with Low-Rank Adaption in Text-to-Image Diffusion Models for High-Definition Synthesis

Diffusion models have revolutionized text-to-image (T2I) synthesis, producing high-quality, photorealistic images. However, they still struggle to properly render the spatial relationships described in text prompts. To address the lack of spatial information in T2I generations, existing methods typically use external network conditioning and predefined layouts, resulting in higher computational costs and reduced flexibility. Our approach builds upon a curated dataset of spatially explicit prompts, meticulously extracted and synthesized from LAION-400M to ensure precise alignment between textual descriptions and spatial layouts. Alongside this dataset, we present ESPLoRA, a flexible fine-tuning framework based on Low-Rank Adaptation, specifically designed to enhance spatial consistency in generative models without increasing generation time or compromising the quality of the outputs. In addition to ESPLoRA, we propose refined evaluation metrics grounded in geometric constraints, capturing 3D spatial relations such as "in front of" or "behind". These metrics also expose spatial biases in T2I models which, even when not fully mitigated, can be strategically exploited by our TORE algorithm to further improve the spatial consistency of generated images. Our method outperforms CoMPaSS, the current baseline framework, on spatial consistency benchmarks.

cs.CV↗

arXiv2024

The role of interface design on prompt-mediated creativity in Generative AI

Generative AI for the creation of images is becoming a staple in the toolkit of digital artists and visual designers. The interaction with these systems is mediated by \emph{prompting}, a process in which users write a short text to describe the desired image's content and style. The study of prompts offers an unprecedented opportunity to gain insight into the process of human creativity. Yet, our understanding of how people use them remains limited. We analyze more than 145,000 prompts from the logs of two Generative AI platforms (Stable Diffusion and Pick-a-Pic) to shed light on how people \emph{explore} new concepts over time, and how their exploration might be influenced by different design choices in human-computer interfaces to Generative AI. We find that users exhibit a tendency towards exploration of new topics over exploitation of concepts visited previously. However, a comparative analysis of the two platforms, which differ both in scope and functionalities, reveals some stark differences. Features diverting user focus from prompting and providing instead shortcuts for quickly generating image variants are associated with a considerable reduction in both exploration of novel concepts and detail in the submitted prompts. These results carry direct implications for the design of human interfaces to Generative AI and raise new questions regarding how the process of prompting should be aided in ways that best support creativity.

cs.CY↗

arXiv2022

Heterogeneous rarity patterns drive price dynamics in NFT collections

We quantify Non Fungible Token (NFT) rarity and investigate how it impacts market behaviour by analysing a dataset of 3.7M transactions collected between January 2018 and June 2022, involving 1.4M NFTs distributed across 410 collections. First, we consider the rarity of an NFT based on the set of human-readable attributes it possesses and show that most collections present heterogeneous rarity patterns, with few rare NFTs and a large number of more common ones. Then, we analyze market performance and show that, on average, rarer NFTs: (i) sell for higher prices, (ii) are traded less frequently, (iii) guarantee higher returns on investment (ROIs), and (iv) are less risky, i.e., less prone to yield negative returns. We anticipate that these findings will be of interest to researchers as well as NFT creators, collectors, and traders.

q-fin.ST↗

arXiv2021

Mapping the NFT revolution: market trends, trade networks and visual features

Non Fungible Tokens (NFTs) are digital assets that represent objects like art, collectible, and in-game items. They are traded online, often with cryptocurrency, and are generally encoded within smart contracts on a blockchain. Public attention towards NFTs has exploded in 2021, when their market has experienced record sales, but little is known about the overall structure and evolution of its market. Here, we analyse data concerning 6.1 million trades of 4.7 million NFTs between June 23, 2017 and April 27, 2021, obtained primarily from Ethereum and WAX blockchains. First, we characterize statistical properties of the market. Second, we build the network of interactions, show that traders typically specialize on NFTs associated with similar objects and form tight clusters with other traders that exchange the same kind of objects. Third, we cluster objects associated to NFTs according to their visual features and show that collections contain visually homogeneous objects. Finally, we investigate the predictability of NFT sales using simple machine learning algorithms and find that sale history and, secondarily, visual features are good predictors for price. We anticipate that these findings will stimulate further research on NFT production, adoption, and trading in different contexts.

q-fin.ST↗

arXiv2021

FairyTailor: A Multimodal Generative Framework for Storytelling

Storytelling is an open-ended task that entails creative thinking and requires a constant flow of ideas. Natural language generation (NLG) for storytelling is especially challenging because it requires the generated text to follow an overall theme while remaining creative and diverse to engage the reader. In this work, we introduce a system and a web-based demo, FairyTailor, for human-in-the-loop visual story co-creation. Users can create a cohesive children's fairytale by weaving generated texts and retrieved images with their input. FairyTailor adds another modality and modifies the text generation process to produce a coherent and creative sequence of text and images. To our knowledge, this is the first dynamic tool for multimodal story generation that allows interactive co-formation of both texts and images. It allows users to give feedback on co-created stories and share their results.

cs.CL↗

arXiv2020

Latent Compass: Creation by Navigation

In Marius von Senden's Space and Sight, a newly sighted blind patient describes the experience of a corner as lemon-like, because corners "prick" sight like lemons prick the tongue. Prickliness, here, is a dimension in the feature space of sensory experience, an effect of the perceived on the perceiver that arises where the two interact. In the account of the newly sighted, an effect familiar from one interaction translates to a novel context. Perception serves as the vehicle for generalization, in that an effect shared across different experiences produces a concrete abstraction grounded in those experiences. Cezanne and the post-impressionists, fluent in the language of experience translation, realized that the way to paint a concrete form that best reflected reality was to paint not what they saw, but what it was like to see. We envision a future of creation using AI where what it is like to see is replicable, transferrable, manipulable - part of the artist's palette that is both grounded in a particular context, and generalizable beyond it. An active line of research maps human-interpretable features onto directions in GAN latent space. Supervised and self-supervised approaches that search for anticipated directions or use off-the-shelf classifiers to drive image manipulation in embedding space are limited in the variety of features they can uncover. Unsupervised approaches that discover useful new directions show that the space of perceptually meaningful directions is nowhere close to being fully mapped. As this space is broad and full of creative potential, we want tools for direction discovery that capture the richness and generalizability of human perception. Our approach puts creators in the discovery loop during real-time tool use, in order to identify directions that are perceptually meaningful to them, and generate interpretable image translations along those directions.

cs.AI↗

arXiv2020

3D Topology Transformation with Generative Adversarial Networks

Generation and transformation of images and videos using artificial intelligence have flourished over the past few years. Yet, there are only a few works aiming to produce creative 3D shapes, such as sculptures. Here we show a novel 3D-to-3D topology transformation method using Generative Adversarial Networks (GAN). We use a modified pix2pix GAN, which we call Vox2Vox, to transform the volumetric style of a 3D object while retaining the original object shape. In particular, we show how to transform 3D models into two new volumetric topologies - the 3D Network and the Ghirigoro. We describe how to use our approach to construct customized 3D representations. We believe that the generated 3D shapes are novel and inspirational. Finally, we compare the results between our approach and a baseline algorithm that directly convert the 3D shapes, without using our GAN.

cs.LG↗

arXiv2020

Covering the News with (AI) Style

We introduce a multi-modal discriminative and generative frame-work capable of assisting humans in producing visual content re-lated to a given theme, starting from a collection of documents(textual, visual, or both). This framework can be used by edit or to generate images for articles, as well as books or music album covers. Motivated by a request from the The New York Times (NYT) seeking help to use AI to create art for their special section on Artificial Intelligence, we demonstrated the application of our system in producing such image.

eess.IV↗

arXiv2014

Prospect Theory for Online Financial Trading

Prospect theory is widely viewed as the best available descriptive model of how people evaluate risk in experimental settings. According to prospect theory, people are risk-averse with respect to gains and risk-seeking with respect to losses, a phenomenon called "loss aversion". Despite of the fact that prospect theory has been well developed in behavioral economics at the theoretical level, there exist very few large-scale empirical studies and most of them have been undertaken with micro-panel data. Here we analyze over 28.5 million trades made by 81.3 thousand traders of an online financial trading community over 28 months, aiming to explore the large-scale empirical aspect of prospect theory. By analyzing and comparing the behavior of winning and losing trades and traders, we find clear evidence of the loss aversion phenomenon, an essence in prospect theory. This work hence demonstrates an unprecedented large-scale empirical evidence of prospect theory, which has immediate implication in financial trading, e.g., developing new trading strategies by minimizing the effect of loss aversion. Moreover, we introduce three risk-adjusted metrics inspired by prospect theory to differentiate winning and losing traders based on their historical trading behavior. This offers us potential opportunities to augment online social trading, where traders are allowed to watch and follow the trading activities of others, by predicting potential winners statistically based on their historical trading behavior rather than their trading performance at any given point in time.

q-fin.TR↗