Neural intel Pod

This document introduces  CodePDE , a new framework for using large language models (LLMs) to generate  code that solves partial differential equations (PDEs) . The authors frame PDE solving as a code generation problem and demonstrate that, with techniques like  debugging and refinement , LLMs can create solvers that are competitive with, and sometimes surpass,  human-written solvers  on various...

This research investigates  online learning for feedforward neural networks  utilizing the sign activation function. The paper identifies a  margin condition  in the first hidden layer as crucial for learnability, demonstrating that the optimal error bound is closely tied to the  totally-separable packing number  of the input space, showing an exponential dependence on dimension in some cases. To...

This document introduces  CityAVOS , a  new benchmark dataset  designed for  Aerial Visual Object Search (AVOS) tasks using  Unmanned Aerial Vehicles (UAVs)  in realistic urban environments. The text describes the unique challenges of urban AVOS, such as complex semantics and difficult identification of similar objects, which differentiate it from previous navigation and object search tasks. It al...

This document introduces  BAT (Benchmark for Auto-bidding Task) , a new resource for researching  autobidding algorithms  in online advertising auctions. It provides a large-scale dataset from the Avito platform, including information on two common auction types,  VCG and FP , along with  campaign details, auction outcomes, and traffic patterns . The paper also presents several  baseline autobiddi...

This collection of texts from Amazon Science highlights the company's extensive research and development efforts across various scientific and technical domains, including  machine learning ,  artificial intelligence , and  robotics . A significant portion focuses on improving the training of large language models, particularly through a novel method called  SeRA which aims to reduce spurious...

This document introduces  T2I-R1 , a novel text-to-image generation model that uses  Reinforcement Learning (RL)  and a  bi-level Chain-of-Thought (CoT)  process to improve image generation. Unlike traditional methods, T2I-R1 leverages  semantic-level CoT  for high-level planning based on the text prompt and  token-level CoT  for detailed, patch-by-patch image generation. A key component is  BiCoT...

This document proposes  pretraining strategies  for estimating  heterogeneous treatment effects (HTE) , particularly the  conditional average treatment effect (CATE) , which varies based on individual characteristics. The core idea is to leverage the  shared support  between factors that are  prognostic  of the outcome (influence baseline risk) and those that are  predictive  of treatment effect h...

This academic paper introduces a novel  mathematical formula  that precisely predicts when a large language model (LLM) might suddenly shift from producing  beneficial output  to generating  incorrect or harmful content , referred to as a "Jekyll-and-Hyde" tipping point. The authors attribute this change to the AI's  attention mechanism , specifically how thinly its attention spreads...

This document introduces Yo’Chameleon, a novel method for personalizing  Large Multimodal Models (LMMs)  like Chameleon. Recognizing that current LMMs lack user-specific knowledge, the paper proposes using only 3-5 images of a  novel concept  to train the model for both  personalized language and vision generation . Yo’Chameleon employs  soft-prompt tuning  to integrate subject-specific informatio...

Three texts published in April 2025 discuss the burgeoning integration and implications of  Artificial Intelligence (AI)  in different sectors. One source focuses on the  financial system , outlining the potential  benefits  and significant  risks  AI presents for financial stability, including operational, market, and cyber concerns, emphasizing the need for monitoring and regulatory adaptation....

The provided research paper investigates  challenges in using process reward models (PRMs) for reinforcement fine-tuning (RFT) of large language models (LLMs) on reasoning tasks , specifically addressing the issue of reward hacking caused by traditional summation-based credit assignment. To mitigate this, the authors introduce  PURE (Process sUpervised Reinforcement lEarning), a novel framework em...

This research explores the use of language models to automate patient record linkage, a crucial process for integrating fragmented healthcare data.  The study investigates the effectiveness of these models for two key tasks:  blocking , which reduces the number of record pairs to compare, and  matching , which determines if two records belong to the same patient. Using real-world cancer registry d...

This survey explores  Parameter-Efficient Continual Fine-Tuning (PECFT) , a method that combines the adaptability of  Continual Learning (CL)  with the efficiency of  Parameter-Efficient Fine-Tuning (PEFT)  to enable large pre-trained models to learn and adapt to new tasks sequentially without forgetting previous knowledge and without extensive retraining. The paper reviews  CL algorithms  and  PE...

Thorsten Ball's article argues that creating a functional code-editing agent is surprisingly straightforward.  It posits that the core components are a large language model (LLM), a loop for interaction, and sufficient tokens. The author demonstrates building a basic agent in Go with fewer than 400 lines of code, showcasing its ability to converse and utilize tools.  The article then introduce...

This research paper introduces a  novel two-stage framework  for more accurately identifying lung tumors in CT scans. The approach first performs a  coarse, full-volume localization  of potential tumors, followed by an  anatomy-aware post-processing  step to refine these regions. Subsequently, a second stage focuses on  detailed segmentation of the identified regions of interest , utilizing uncert...

This research introduces  Complex-Edit , a new benchmark for evaluating how well image editing models follow instructions with varying levels of complexity. The benchmark was created using  GPT-4o  to generate atomic editing tasks, which were then simplified and combined into more intricate instructions. The authors also present  a suite of metrics and a VLM-based evaluation system  to assess inst...

This research introduces " sleep-time compute ," a novel method for enhancing large language model efficiency by allowing them to process contextual information offline, before user queries arrive. By  anticipating potential questions and pre-computing relevant inferences , this approach significantly  reduces the computational resources and latency needed at test time  to achieve compar...

This paper introduces Miras, a novel framework for designing sequence models by drawing parallels between neural architectures and associative memory.  The authors reconceptualize models like Transformers and linear RNNs as associative memory modules guided by an "attentional bias" objective.  Miras offers a unified perspective based on four key choices: memory architecture, attentional...

This paper introduces  RUKA , a newly designed  open-source, low-cost, and anthropomorphic robotic hand  intended for research in dexterous manipulation. It details  RUKA's hardware design , emphasizing its tendon-driven actuation, 3D-printed components, and off-the-shelf parts, achieving a balance of  precision, durability, and affordability . The paper further presents a  data-driven control...

The provided paper introduces  GenEAva , a novel framework for generating high-quality cartoon avatars with detailed facial expressions by fine-tuning a diffusion model on realistic faces and then applying stylization. To overcome limitations in existing datasets, the authors created  GenEAva 1.0 , the first expressive cartoon avatar dataset featuring 13,230 avatars across 135 distinct expressions...

The provided text introduces  VCR-Bench , a novel benchmark designed to evaluate the  Chain-of-Thought (CoT) reasoning capabilities  of large vision-language models (LVLMs) in video understanding. Current benchmarks for video understanding often fall short by  not thoroughly assessing the reasoning process , focusing mainly on final answer accuracy and struggling to differentiate between perceptio...

This Amazon Science blog post introduces  HalluMeasure , a novel approach for  automatically detecting hallucinations in large language model outputs. The method employs a  three-pronged strategy  combining claim-level evaluations, chain-of-thought reasoning, and the classification of hallucination error types. By decomposing LLM responses into individual claims and comparing them to relevant cont...

Meta AI has introduced its Llama 4 family of large language models , highlighting two new open-weight models:  Llama 4 Scout and Llama 4 Maverick , along with a powerful, still-training model called  Llama 4 Behemoth .  These multimodal models feature innovations like Mixture-of-Experts architecture and significantly expanded context windows , outperforming previous Llama iterations and competing...

This research introduces  DISCIPL , a novel framework where language models dictate their own reasoning process for complex tasks. By having a  Planner LM  generate inference programs, DISCIPL orchestrates  Follower LMs  to solve problems, offering a more efficient and automated approach compared to existing methods like chain-of-thought or structured inference. The framework supports various infe...

The provided text introduces  Amazon Q Developer in SageMaker Canvas , a new AI tool that allows users to build and deploy machine learning models using natural language. This tool, powered by generative AI, simplifies the ML development process by automating tasks like data preparation, model selection, and training, thus requiring no specialized expertise. The article details the functionality o...