AI/ML arXiv cs.AI

Enhancing Generative Auto-bidding with Offline Reward Evaluation and Policy Search

Presents AIGB-Pearl, a method integrating generative planning and policy optimization to enhance auto-bidding in advertising systems.

AI/ML arXiv cs.AI

RoboSSM: Scalable In-context Imitation Learning via State-Space Models

Introduces RoboSSM, a scalable in-context imitation learning framework for robots using state-space models (Longhorn) instead of Transformers.

AI/ML arXiv cs.AI

MENTOR: Reinforcement Learning via Flexible Teacher-Optimized Rewards for Tool-Use Distillation

Proposes MENTOR, a RL-based approach to distill tool-use capabilities from LLMs into smaller models using flexible teacher-optimized rewards.

AI/ML arXiv cs.AI

Beyond Reasoning Gains: Mitigating General-Capability Forgetting in Large Reasoning Models

Introduces RECAP, a replay strategy with dynamic objective reweighting to prevent capability forgetting in large reasoning models during post-training.

AI/ML arXiv cs.AI

Bid Farewell to Seesaw: Towards Accurate Long-tail Session-based Recommendation via Dual Constraints of Hybrid Intents

Proposes the HID framework to improve long-tail session-based recommendations by reducing the noise associated with low-exposure items.

Software Engineering Hacker News

Hey, N00B, We Didn't Hire You to Complete Tasks

A discussion on the expectations for junior developers, arguing that their role is more than just completing assigned tasks.

Software Engineering Hacker News

Digital Printing of Arabic: explaining the problem

An explanation of the technical challenges involved in the digital printing of Arabic text.

Open Source TechCrunch

He made your free video player run smoothly. Now he’s doing that for robots.

Jean-Baptiste Kempf, creator of VLC, is developing Kyber, an infrastructure layer for real-time remote device control.

AI/ML arXiv cs.AI

A Deep Generative Model for Resting-State EEG Synthesis and Transferable Representation Learning

REST-GAN is a GAN-based framework for synthesizing resting-state EEG signals and unsupervised feature extraction.

AI/ML arXiv cs.AI

TerraMind: Large-Scale Generative Multimodality for Earth Observation

TerraMind is an any-to-any generative multimodal foundation model designed for Earth observation geospatial data.

AI/ML arXiv cs.AI

Bridging Distribution Shift and AI Safety: Conceptual and Methodological Synergies

A conceptual analysis exploring the synergies between distribution shift and AI safety research.

AI/ML arXiv cs.AI

Overcoming Labelled Data Scarcity for Defect Classification in Scanning Tunneling Microscopy

A new automated approach for segmenting STM images using few-shot and unsupervised learning to reduce the need for labeled data.

AI/ML arXiv cs.AI

Critique of World Model

A critique of the 'World Model' concept and the proposal of a new Generative Latent Prediction (GLP) architecture for AGI.

AI/ML arXiv cs.AI

Assessment of Personality Dimensions Across Situations in Dyadic Role-Play Scenarios

Research on how perceived personality traits in AI assistants can be predicted using acoustic and non-verbal features across different situations.

AI/ML arXiv cs.AI

On the Limitations of Ray-Tracing for Learning-Based RF Tasks in Urban Environments

An evaluation of the limitations of ray-tracing in Sionna for learning-based RF tasks in urban environments.

Other Hacker News

Iran requires insurance on ships using Strait of Hormuz, fees likely to follow

Iran is requiring insurance for ships using the Strait of Hormuz, which is expected to lead to increased fees.

AI/ML arXiv cs.AI

ScaleWoB: Guiding GUI Agents with Coding Agents via Large-Scale Environmental Synthesis

ScaleWoB is a framework for creating high-fidelity synthetic interactive environments to evaluate and train GUI agents across mobile, desktop, and automotive platforms.

AI/ML arXiv cs.AI

FundaPod: A Multi-Persona Agent Pod Platform with Knowledge Graph Memory for AI-Assisted Fundamental Investment Research

FundaPod is a multi-persona agent platform designed for fundamental investment research, using a knowledge-graph memory system to surface disagreements between AI agents for human adjudication.

AI/ML arXiv cs.AI

VitalAgent: A Tool-Augmented Agent for Reactive and Proactive Physiological Monitoring over Wearable Health Data

VitalAgent is a tool-augmented agentic framework for reactive and proactive physiological monitoring of ECG/PPG data from wearables, accompanied by the VitalBench benchmark dataset.

AI/ML arXiv cs.AI

Science Earth: Towards A Planet-Scale Operating System for AI-Native Scientific Discovery

Science Earth proposes a planet-scale scientific runtime that allows diverse AI capabilities and lab hardware to connect and collaborate dynamically via the EACN protocol for scientific discovery.