paperlist.ai

AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

Resource-Efficient RGB-Only Action Recognition for Edge Deployment

Dongsik Yoon, Jongeun Kim, Dayeon Lee

Resource-constrained assistive monitoring requires compact local video perception and an explicit understanding of how recognition reliability changes under common visual degradation. We present a compact RGB-only action-recognition network that combines an X3D-style hierarchy with temporal shift, a 3D Universal Inverted Bottleneck, Ghost pointwise convolutions, factorized depthwise operators, selective temporal adaptation, and parameter-free attention. The model attains 95.10/98.31% on NTU RGB+D 60 (X-Sub/X-View) and 90.88/92.66% on NTU RGB+D 120 (X-Sub/X-Set) with only 0.96-0.99M parameters. To connect compact deployment with assistive use, we evaluate frozen checkpoints without retraining on the datasets' official Medical Conditions groups while retaining the original 60-/120-way decision spaces. Medical macro recall reaches 93.19/97.93% on NTU60 and 91.47/91.48% on NTU120. Under controlled Level-2 visual stress on NTU60 X-Sub, medical recall drops by 1.77 points under low light, 2.23 under motion blur, 4.87 under crop perturbation, and 19.05 under occlusion; among the four selected Level-2 perturbations, occlusion causes the largest observed degradation. On Jetson Orin Nano, the TensorRT FP16 engine occupies 5.30 MiB, demonstrating strong static compactness but not throughput superiority. The results position the model as a compact perception component for lower-rate assistive monitoring, rather than as a complete clinical or robotic system.

arXiv

Where Does Vision Meet Language? Understanding and Refining Visual Fusion in MLLMs via Contrastive Attention

Shezheng Song, Shasha Li, Shan Zhao, Xiaopeng Li, Qian Wan, Chengyu Wang, Tianwei Yan, Jun Ma, Jie Yu

Multimodal Large Language Models (MLLMs) have achieved remarkable progress in vision-language understanding, yet how they internally integrate visual and textual information remains poorly understood. To bridge this gap, we perform a systematic layer-wise masking analysis across multiple architectures, revealing how visual-text fusion evolves within MLLMs. The results show that fusion emerges at several specific layers rather than being uniformly distributed across the network, and certain models exhibit a late-stage "review" phenomenon where visual signals are reactivated before output generation. Besides, we further analyze layer-wise attention evolution and observe persistent high-attention noise on irrelevant regions, along with gradually increasing attention on text-aligned areas. Guided by these insights, we introduce a training-free contrastive attention framework that models the transformation between early fusion and final layers to highlight meaningful attention shifts. Extensive experiments across various MLLMs and benchmarks validate our analysis and demonstrate that the proposed approach improves multimodal reasoning performance.

arXiv

Edit2Restore:Few-Shot Image Restoration via Parameter-Efficient Adaptation of Pre-trained Editing Models

Mustafa Akın Yılmaz, Ahmet Bilican, Burak Can Biner, Ahmet Murat Tekalp

Image restoration has traditionally required training specialized models on thousands of paired examples per degradation type. Large pre-trained text-conditioned image editing models encode rich priors about image structure, quality, and degradation, yet we find that this knowledge does not, on its own, make them restorers: state-of-the-art editing models largely fail at restoration in the zero-shot regime. We show that what these priors lack is not capability but direction, and that a small amount of parameter-efficient adaptation supplies it. Fine-tuning LoRA adapters on FLUX.1 Kontext, a 12B-parameter flow matching model for image-to-image translation, with only 32--128 paired images per task and guided by simple text prompts, we turn a mediocre zero-shot editor into a competitive restorer. A single unified adapter, conditioned on task-specific prompts, handles five diverse degradations. Despite using three to four orders of magnitude less data, our few-shot model surpasses a recent restoration baseline trained on over a million curated pairs on the majority of perceptual and distribution-level metrics, on which we evaluate in keeping with our focus on perceptual rather than pixel-fidelity quality. Through comprehensive studies, we analyze the impact of training-set size, the trade-off between task-specific and unified multi-task adapters, the effect of text encoder adaptation, and zero-shot baseline performance, establishing pre-trained editing models as a compelling, data-efficient foundation for few-shot, prompt-guided image restoration.

arXiv

INFORM-CT: INtegrating LLMs and VLMs FOR Incidental Findings Management in Abdominal CT

Idan Tankel, Nir Mazor, Rafi Brada, Christina LeBedis, Guy ben-Yosef

Incidental findings in CT scans, though often benign, can have significant clinical implications and should be reported following established guidelines. Traditional manual inspection by radiologists is time-consuming and variable. This paper proposes a novel framework that leverages large language models (LLMs) and foundational vision-language models (VLMs) in a plan-and-execute agentic approach to improve the efficiency and precision of incidental findings detection, classification, and reporting for abdominal CT scans. Given medical guidelines for abdominal organs, the process of managing incidental findings is automated through a planner-executor framework. The planner, based on LLM, generates Python scripts using predefined base functions, while the executor runs these scripts to perform the necessary checks and detections, via VLMs, segmentation models, and image processing subroutines. We demonstrate the effectiveness of our approach through experiments on a CT abdominal benchmark for three organs, in a fully automatic end-to-end manner. Our results show that the proposed framework outperforms existing pure VLM-based approaches in terms of accuracy and efficiency.

arXiv

Redefining Generalization in Visual Domains: A Two-Axis Framework for Fake Image Detection with FusionDetect

Amirtaha Amanzadi, Zahra Dehghanian, Hamid Beigy, Hamid R. Rabiee

The rapid development of generative models has made it increasingly crucial to develop detectors that can reliably detect synthetic images. Although most of the work has now focused on cross-generator generalization, we argue that this viewpoint is too limited. Detecting synthetic images involves another equally important challenge: generalization across visual domains. To bridge this gap,we present the OmniGen Benchmark. This comprehensive evaluation dataset incorporates 12 state-of-the-art generators, providing a more realistic way of evaluating detector performance under realistic conditions. In addition, we introduce a new method, FusionDetect, aimed at addressing both vectors of generalization. FusionDetect draws on the benefits of two frozen foundation models: CLIP & Dinov2. By deriving features from both complementary models,we develop a cohesive feature space that naturally adapts to changes in both thecontent and design of the generator. Our extensive experiments demonstrate that FusionDetect delivers not only a new state-of-the-art, which is 3.87% more accurate than its closest competitor and 6.13% more precise on average on established benchmarks, but also achieves a 4.48% increase in accuracy on OmniGen,along with exceptional robustness to common image perturbations. We introduce not only a top-performing detector, but also a new benchmark and framework for furthering universal AI image detection. The code and dataset are available at http://github.com/amir-aman/FusionDetect

arXiv

Concept Unlearning by Modeling Key Steps of Diffusion Process

Chaoshuo Zhang, Chenhao Lin, Zhengyu Zhao, Le Yang, Qian Wang, Chao Shen

Text-to-image diffusion models remain susceptible to generating undesirable or harmful content. Although concept unlearning mitigates this risk, existing methods struggle with a critical optimization dilemma: thorough semantic erasure frequently induces the catastrophic forgetting of unrelated generative capabilities. To overcome this challenge, we propose Key Step Concept Unlearning (KSCU). Serving as an integrated methodological refinement deeply motivated by information theory, KSCU explores the profound impact of step scheduling order and reveals that traditional randomized timestep sampling severely disrupts trajectory dependency. We demonstrate that indiscriminately targeting the entire diffusion process is inefficient, as the optimal step range for unlearning inherently varies across different concepts. Rather than globally fine-tuning all timesteps, KSCU explicitly integrates a sequential-scheduling-based Key Step Table, CFG-aware leakage compensation, and prompt augmentation to dynamically isolate optimization to a concept-specific active region. This localized strategy successfully eradicates the target concept while preventing the structural collapse caused by early-step over-optimization. Consequently, KSCU significantly reduces computational overhead and establishes a state-of-the-art trade-off between concept erasure and utility retention. Comprehensive evaluations demonstrate that KSCU consistently delivers superior performance across diverse unlearning tasks, including nudity, style, object classes, and mass instance concepts. For example, in nudity removal, KSCU yields a 96.5% unlearning accuracy alongside a state-of-the-art FID of 14.1.

arXiv

CoCA: Step-level Reward for Free in RL-based T2I Diffusion Model Fine-tuning

Xinyao Liao, Wei Wei, Xiaoye Qu, Qiyuan He, Angela Yao, Yu Cheng

Recent advances in text-to-image (T2I) diffusion model fine-tuning leverage reinforcement learning (RL) to align generated images with learnable reward functions. The existing approaches reformulate denoising as a Markov decision process for RL-driven optimization. However, they suffer from reward sparsity, receiving only a single delayed reward per generated trajectory. This flaw hinders precise step-level attribution of denoising actions, undermines training efficiency. To address this, we propose a simple yet effective credit assignment framework that dynamically distributes dense rewards across denoising steps. Specifically, we track changes in cosine similarity between intermediate and final images to quantify each step's contribution on progressively reducing the distance to the final image. Our approach avoids additional auxiliary neural networks for step-level preference modeling and instead uses reward shaping to highlight denoising phases that have a greater impact on image quality. Our method achieves 1.25 to 2 times higher sample efficiency and better generalization across four human preference reward functions, without compromising the original optimal policy.

arXiv

Tracking the Flight: Exploring a Computational Framework for Analyzing Escape Responses in Plains Zebra (Equus quagga)

Isla Duporge, Sofia Minano, Nikoloz Sirmpilatze, Igor Tatarnikov, Scott Wolf, Adam L. Tyson, Daniel Rubenstein

Ethological research increasingly benefits from the growing affordability and accessibility of drones, which enable the capture of high-resolution footage of animal movement at fine spatial and temporal scales. However, analyzing such footage presents the technical challenge of separating animal movement from drone motion. While non-trivial, computer vision techniques such as image registration and Structure-from-Motion (SfM) offer practical solutions. For conservationists, open-source tools that are user-friendly, require minimal setup, and deliver timely results are especially valuable for efficient data interpretation. This study evaluates three approaches: a bioimaging-based registration technique, an SfM pipeline, and a hybrid interpolation method. We apply these to a recorded escape event involving 44 plains zebras, captured in a single drone video. Using the best-performing method, we extract individual trajectories and identify key behavioral patterns: increased alignment (polarization) during escape, a brief widening of spacing just before stopping, and tighter coordination near the group's center. These insights highlight the method's effectiveness and its potential to scale to larger datasets, contributing to broader investigations of collective animal behavior.

arXiv

HierDAMap: Towards Universal Domain Adaptive BEV Mapping via Hierarchical Perspective Priors

Siyu Li, Yihong Cao, Hao Shi, Yongsheng Zang, Xuan He, Kailun Yang, Zhiyong Li

The exploration of Bird's-Eye View (BEV) mapping technology has driven significant innovation in visual perception technology for autonomous driving. BEV mapping models need to be applied to the unlabeled real world, making the study of unsupervised domain adaptation models an essential path. However, research on unsupervised domain adaptation for BEV mapping remains limited and cannot perfectly accommodate all BEV mapping tasks. To address this gap, this paper proposes HierDAMap, a universal and holistic BEV domain adaptation framework with hierarchical perspective priors. Unlike existing research that solely focuses on image-level learning using prior knowledge, this paper explores the guiding role of perspective prior knowledge across three distinct levels: global, sparse, and instance levels. With these priors, HierDAMap consists of three essential components, including Semantic-Guided Pseudo Supervision (SGPS), Dynamic-Aware Coherence Learning (DACL), and Cross-Domain Frustum Mixing (CDFM). SGPS constrains the cross-domain consistency of perspective feature distribution through pseudo labels generated by vision foundation models in 2D space. To mitigate feature distribution discrepancies caused by spatial variations, DACL employs uncertainty-aware predicted depth as an intermediary to derive dynamic BEV labels from perspective pseudo-labels, thereby constraining the coarse BEV features derived from corresponding perspective features. CDFM, on the other hand, leverages perspective masks of the view frustum to mix multi-view perspective images from both domains, which guides cross-domain view transformation and encoding learning through mixed BEV labels. Furthermore, this paper introduces intra-domain feature exchange data augmentation to enhance the efficiency of domain adaptation learning. The source code will be made publicly available at https://github.com/lynn-yu/HierDAMap.

arXiv

Ordinal-Aware Calibration for Ordinal Classification

Daehwan Kim, Haejun Chung, Ikbeom Jang

Deep neural networks frequently produce overconfident, miscalibrated predictions. In ordinal classification, predictions must also adhere to a unimodal and order-consistent structure, a requirement that has dominated prior work while overlooking calibration. We formalize this joint challenge as ordinal calibration for the first time and propose the Ordinal loss for Calibration and Unimodality (ORCU). Unlike incremental modular combinations, ORCU provides a concise, principled unification of distance-aware soft encoding and an ordinal-aware log-barrier extension on adjacent logit gaps. This coupling is theoretically synergetic: soft-encoded targets pre-condition the optimization landscape by providing a finite, full-support ordinal anchor for well-posed refinement, while the log-barrier extension preserves non-vanishing adjacent-gap gradients in the unimodality-feasible interior. This enables the log-barrier to refine confidence without disrupting the ordinal geometry. ORCU requires no architectural changes or post-hoc calibration. Across four ordinal benchmarks, it achieves state-of-the-art calibration without compromising accuracy. We establish a new reliability standard by providing a reproducible benchmark over 10 specialized and general-purpose loss functions. Our code is available at https://github.com/labhai/ORCU.

arXiv

OTIS: Learning High-Quality Time Series Features With Tiny Encoders

Özgün Turgut, Philip Müller, Martin J. Menten, Daniel Rueckert

We introduce OTIS, an open time series encoder that yields high-quality time series features for downstream deployment on any system, including resource-constrained wearables and industrial sensors. Currently, the development of powerful general-purpose encoders relies on the scaling laws hypothesis, using large encoder sizes to memorise the heterogeneous distributions of multi-domain training data. However, this reliance on scale creates a barrier to real-world utility, rendering deployment on resource-constrained systems infeasible due to strict memory, energy, and latency constraints. Surprisingly, we find that tailoring standard masked modelling pre-training to time series properties yields a tiny $7.1\,$M encoder that matches the state-of-the-art performance of $54\times$ larger encoders across $162$ tasks, while requiring $10\times$ less memory, $43\times$ less energy, and $37\times$ lower latency. To achieve this without the capacity tax, we introduce three novel components: (1) a domain-aware tokeniser to resolve conflicting semantics within multi-domain training data; (2) a dual masking strategy to capture spatiotemporal structures and temporal causality; and (3) a structure-aware objective to decouple feature learning from modelling noise. Consequently, OTIS produces high-quality time series features that enable state-of-the art performance in discriminative tasks and even extend seamlessly to generative tasks at minimal additional cost. To democratise access to powerful time series features on any system, we release our code and pre-trained weights.

arXiv

Logarithmic Mathematical Morphology: Theory and applications

Guillaume Noyel

In Mathematical Morphology for grey-level functions, an image is analysed by another image named the structuring function. This structuring function is translated over the image domain and summed to the image. However, in an image presenting lighting variations, when the structuring function is added to this image, the amplitude of this translated structuring function should vary according to the image intensity. Such a property is not verified in Mathematical Morphology for grey level functions, when the structuring function is summed to the image with the usual additive law. This issue has been addressed by defining a framework that uses an additive law, for which the amplitude of the structuring function added to the image varies according to the amplitude of the image itself. This additive law is chosen within the Logarithmic Image Processing framework and models the lighting variations with a physical cause such as a change of light intensity. The framework is named Logarithmic Mathematical Morphology (LMM) and allows the definition of operators which are robust to such lighting variations.

arXiv

MicroPython and CircuitPython: Pythons Quiet Takeover of IoT and Robotics

Sayed Mahbub Hasan Amiri, Atiar Zahan

Background: Python has become the dominant language in software and data science, yet embedded systems have remained tied to C/C++ due to performance and memory constraints. MicroPython and CircuitPython are changing this by bringing Python to microcontrollers, lowering barriers for IoT and robotics development. Aim: This article examines whether these platforms are achieving a quiet takeover of embedded systems, focusing on ecosystem growth, practical applications, performance trade-offs, educational adoption, and prospects. Methods: A mixed-methods design was used, including quantitative analysis of GitHub, Stack Overflow, and Google Trends data; curation of case studies from Hackster.io, Hackaday.io, and the Adafruit Learning System; and original benchmarks on ESP32 and Raspberry Pi Pico comparing MicroPython, CircuitPython, and Arduino C++ across GPIO, I2C, SPI, Wi-Fi, and memory usage. Results: Metrics show sustained growth, with MicroPython supporting over 200 boards and CircuitPython over 400. Benchmarks reveal 10-20 times slower I/O and four to six times higher memory use than C, but performance remains adequate for common sensor and network tasks. Case studies demonstrate successful deployment in home automation, robotics, wearables, agriculture, and professional prototyping. Education emerges as a primary adoption driver. Conclusions: Python is not replacing C/C++; rather, it is becoming the default prototyping and educational language for embedded systems. Continued hardware improvements, better tooling, and standardization will likely deepen this trend. The article offers balanced, evidence-based insights for developers, educators, and technology decision-makers.

arXiv

Trust Is Not Enough: Influence Calibration for On-Policy Self-Distillation in Agentic RL

Qizhen Lan, Xi Xiao, Xiangchen Guan, Mengchen Fan, Moule Lin, Jung Im Choi, Lijing Zhu

On-policy self-distillation (OPSD) gives language agents dense token-level supervision from a privileged self-teacher on the policy's own trajectories. Existing methods allocate this supervision mainly by teacher trust, but trust does not reveal whether emphasizing a token supports the current policy objective. We call this the trust-utility mismatch and introduce Influence Calibration for Self-Distillation (ICSD). For each supervised token, ICSD measures the first-order response of its importance-weighted RL surrogate contribution to a teacher-directed output perturbation. Batch-adaptive calibration converts this non-stationary signal into a bounded allocation weight while preserving the original auxiliary-loss mass within each action turn. These detached weights affect only the distillation loss and require no additional model pass. Across ALFWorld, WebShop, and Search-QA, ICSD improves all matched aggregate metrics over trust-only allocation under Group Relative Policy Optimization (GRPO) and Group-in-Group Policy Optimization (GiGPO), across two model families spanning 1.5B to 7B. At 7B, it reaches 96.1% ALFWorld success and a WebShop score of 93.1. Frozen-batch analyses show that ICSD reduces teacher-supported mass assigned to objective-opposed tokens from 60.1% to 37.8% and raises cosine compatibility with the RL gradient by 0.192. A companion repository is avail- able at https://github.com/lanqz7766/Influence-Calibration-for-On-Policy-Self-Distillation-in-Agentic-RL.

arXiv

SkillComposer: Learning Reusable Skills for Natural-Language Robot Programming

John Woods, Hasti Seifi

Natural-language interfaces can lower the barrier to programming robots, but existing systems struggle when users request complex tasks. While large language models (LLMs) perform well with simple commands, they often struggle to generate code for multi-step tasks, decompose high-level instructions, or reuse prior solutions. We present SkillComposer, an interactive natural-language robot programming system for simulation environments that continually learns reusable program abstractions. SkillComposer uses a generate-test architecture in which an LLM iteratively generates and revises robot programs before execution. Successful programs are stored and processed by an online library-learning algorithm that compresses recurring function sequences into reusable macro skills for future tasks. We evaluate SkillComposer through ablation experiments and a user study with 12 participants to determine its effectiveness on manipulation and robot caregiving tasks. The results show that evaluator-guided generation and learned abstractions improve success rates and usability while reducing user effort in natural-language robot programming.

arXiv

Training Leaves Traces: Centered Residual Signatures for Language Model Lineage Verification

Aman Singh Thakur, Rayan Khoury

Open-weight language models are fine-tuned, quantized, pruned, and merged, yet their provenance is often undocumented. We study data-free white-box lineage verification: can weights alone reveal whether two compatible model checkpoints share ancestry? Residual training produces a shared identity-aligned component in branch products, so this structure alone cannot establish ancestry. We remove it and compare checkpoint-specific structure across residual blocks, yielding a symmetric lineage score calibrated against independent checkpoints. On residual-MLP and GPT-2 benchmarks, the score separates fine-tuned, LoRA-merged, pruned, and quantized descendants from independent and distilled models (AUROC=1.0), distinguishing weight ancestry from behavioral similarity. Under function-preserving checkpoint laundering experiments, weight-space baselines lose margin or fail; our score remains unchanged and runs 76x faster than the nearest robust baseline on GPT-2. The projection-pairing signal appears across six language-model families and beyond, and a case study correctly identifies 3 related and 7 unrelated LLaMA-2 public checkpoints. Collectively, these results establish a passive, data-free provenance signal for compatible open-weight language-model checkpoints

arXiv

LLMs Can Predict Failure Risk, But Struggle to Predict Which Collaboration Protocol Pays Off: Cost-Aware Protocol Routing Across Reasoning Tasks

Chih-Hsuan Yang, Jingyan Jiang, Cheng-Hau Yang, Vikram Vasudevan, Huihuo Zheng, Venkatram Vishwanath, Rajeev Thakur

Multi-agent large language model (LLM) systems can improve reasoning by spending more computation, but deployment requires deciding when extra collaboration is worth its cost. We isolate this decision by running every problem under four protocols while holding the solver fixed within each setting: direct solving (Baseline), iterative self-correction (Single), planner-executor-reviewer collaboration (PER), and multi-agent deliberation (Broadcast). The primary benchmark comprises 4,181 competition-level math problems; paired robustness checks cover four benchmarks spanning competition math, biology, and broader science with two solver families. Across fixed policies, trained routers, and frozen LLM routers, conservative policies under-escalate, whereas higher-solve frozen routers often over-escalate. A post-answer, pre-collaboration gpt-oss-120b probe ranks Baseline failures with 0.8847 AUROC (4,151 parseable cases; 95% CI [0.8732, 0.8955]). The same score remains informative for predicting whether any collaboration helps (0.7683 AUPRC), but is much weaker for identifying PER- or Broadcast-specific value (0.1674 and 0.1041 AUPRC). Separately, the pre-answer self-confidence gate reaches 78.0% solve at 45K tokens, compared with 73.8% at 71.3K for a frozen gpt-oss-120b router and 92.4% for a retrospective fixed-order oracle. Across 10 paired model-condition settings, the oracle adds 23.2-58.3 points of retrospective coverage over Baseline, but protocol profiles vary by task. In the six settings with held-out router evaluations, oracle gaps remain 18.5-28.9 points. Confidence can therefore support initial escalation, while protocol-specific cost-aware routing remains unresolved.

arXiv

Optimal Watermark Localization in Mixed-Source Large Language Model Texts

Jose H. Blanchet, T. Tony Cai, Xiang Li, Hao Liu, Qi Long, Weijie J Su

Watermarking provides a principled way to authenticate text generated by large language models (LLMs). In practice, however, the final text may be mixed-source, with watermark evidence surviving at only a subset of token positions after rewriting, insertion, deletion, or paraphrasing. Although prior work has studied global detection of watermark signals, when such signals can be localized remains unclear. We formulate watermark localization as a token-level multiple-testing problem based on pivotal statistics, with a latent indicator recording whether watermark dependence survives at each position. Under an asymptotic regime indexed by exponents for signal sparsity, next-token concentration, and effective-vocabulary growth, we derive a sharp boundary for global detection and phase transitions for discovery and classification within the class of coordinatewise pivot-based localization rules. We show that discovery is strictly harder than detection and that consistent classification is impossible across the parameter regime within this class. We then develop an adaptive thresholding method that does not require knowledge of the exponents or time-varying next-token distributions, but uses a data-driven estimate of the surviving watermark fraction. The method attains the optimal discovery boundary and near-optimal discovery power relative to homogeneous pivot-based rules. Simulations support the theoretical phase transitions, while experiments on model-generated texts demonstrate practical localization performance under common edit mechanisms.

arXiv

Interpretable Cross-Lingual Alignment in Small Language Models: Probing Cultural and Pragmatic Reasoning in Japanese-English Bilingual LLMs

Florian Braun

Large language models work well on English and behave in poorly understood ways on languages typologically far from it. Japanese is a clean example, where evaluation still leans on translation quality and JGLUE-style benchmarks, which roll lexical, syntactic and pragmatic competence into a single score. The phenomena on which general-purpose models fail Japanese users are pragmatic: honorifics, in-group and out-group reference, context-sensitive politeness, zero anaphora. I introduce J-PragEval-v0, a minimal-pair benchmark isolating four such phenomena from surface fluency, and combine it with linear probes and teacher-forced log-probability evaluation to ask where inside TinySwallow-1.5B (28 layers, hidden size 1536) the corresponding contrasts live. The four features split three ways. Honorific register sits cleanly in the residual stream: 0.96 balanced accuracy at layer 15, and the model flips its preferred continuation with the scenario on 93 percent of items. Implicit subject and in-group reference are not linearly decodable at the final prompt token (0.48 and 0.38), yet flip rates are 0.77 and 0.79, so the contrast is worked out during generation rather than stored at the prompt. Indirect refusal is the negative case: 0.95 probe accuracy collapsing to a 0.43 flip rate under length-normalised teacher forcing, because the current minimal pairs conflate politeness with continuation length. I also specify Pragmatic Representation Steering, a parameter-free inference-time method that edits residual-stream activations along the class-mean-difference directions probing identifies. Feasibility is argued indirectly rather than demonstrated: the contrastive activation addition baseline, the same geometry the method would inject, recovers probe accuracy within one to two points of logistic regression wherever a linear signal exists. Scaling to Llama-3.1-Swallow-8B is the next step.

arXiv

Where Does Retrieval Fail? Evaluating RAG Architectures for Agricultural Advisory

Khan Raiyan Ibne Reza, Sanjana Aktar Maria, Sumaiya Tabassum Nimi

Retrieval quality in RAG systems is commonly reported as a single aggregate score, which can hide large differences across query types and language conditions. We study this problem in Bengali agricultural advisory, where farmer queries are often colloquial while official advisory documents use formal scientific terminology. We construct a test collection of 1,000 queries and 2,882 knowledge nodes extracted from 284 official Bangladeshi agricultural publications, and use it to evaluate five retrieval architectures and six embedding models under three controlled language conditions. The results show that no single retrieval method is consistently best. For native Bengali queries, BM25 is the strongest single retriever (R@10 = 0.506) while Hybrid RRF reaches the highest overall R@10 of 0.539. However, dense retrieval performance varies sharply by query type: R@10 is 0.093 on colloquial farmer queries and 0.970 on formal safety queries. Across language conditions, BM25 R@10 drops from 0.506 on Bengali queries to 0.004 when English queries are matched against the Bengali corpus, while dense retrieval falls only from 0.464 to 0.425. We also find that embedding task configuration and passage length can each change reported R@10 by a factor of seven, independent of architecture. These results show why low-resource RAG evaluation should report performance by language condition and query type rather than relying on aggregate scores alone. The dataset and evaluation scripts are available at https://huggingface.co/datasets/RaiyanKhaan/AgriTrust-RAG.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.