The Hidden Geometry of Deep Learning
![The simplex optimization method, while appearing geometrically straightforward, subtly introduces an implicit bias toward solutions favoring larger steps - a phenomenon evidenced by its tendency to converge more rapidly along axes than within polyhedral facets, effectively prioritizing speed over a truly exhaustive search of the solution space, as described by [latex] \nabla f(x) [/latex].](https://arxiv.org/html/2603.02622v1/2603.02622v1/ImplicitBias.png)
New research reveals how optimization algorithms in deep linear discriminant analysis subtly enforce geometric constraints, impacting model behavior.
Science
Beyond Text: Tapping Hidden Numerical Insights in Language Models
New research reveals that large language models possess surprisingly accurate numerical prediction capabilities encoded within their internal states, bypassing the need for traditional text generation.
Smart Ray Tracing: Machine Learning Accelerates Radio Wave Simulation
A new machine learning framework intelligently samples radio propagation paths, dramatically speeding up simulations while maintaining accuracy.
Beyond Homophily: A Transformer Approach to Graph Fraud Detection
A new model leverages multi-scale neighborhood awareness to improve the detection of fraudulent activity in complex network data.
Markets React: How Political Shocks Ripple Through Election Predictions
![Trading activity in digitally-created “Biden Tokens” spiked around key political events, demonstrating how speculative fervor transforms symbolic capital into quantifiable market behavior and revealing the susceptibility of even novelty assets to emotionally-driven surges and declines-a pattern mirroring established financial bubbles driven by hope and fear rather than underlying value, as predictably encoded in [latex] P = f(E, S) [/latex], where price (P) is a function of events (E) and sentiment (S).](https://arxiv.org/html/2603.03152v1/2603.03152v1/biden_no_token_price_tx_vol_events.png)
New research examines how prediction markets process unexpected political events, revealing the interplay between shifting beliefs and immediate trading pressures.
Predicting the Future: Inside Polymarket’s 2024 Election Bets
![The shifting correlation between predictions in the Trump and Democrat-aligned forecasting markets demonstrates a dynamic relationship, revealing how consensus diverges and converges as events unfold-a pattern indicative of evolving perceptions rather than static alignment, and suggesting that even opposing systems momentarily share predictive signals before ultimately re-establishing independent trajectories-a phenomenon mirroring the inherent impermanence of all complex systems [latex] \Delta t \rightarrow \in fty [/latex].](https://arxiv.org/html/2603.03136v1/2603.03136v1/rolling_correlation_between_trump_and_dem_market.png)
A detailed analysis of transaction data from the blockchain-based prediction market Polymarket reveals how it evolved during the 2024 presidential election cycle.
Beyond Accuracy: How Optimizer Choice Shapes Financial Forecasts
![Model sensitivity, as measured by impulse responses at [latex]t-1[/latex], diverges across different optimization algorithms, highlighting the nuanced impact of each on system dynamics.](https://arxiv.org/html/2603.02620v1/2603.02620v1/figs/difference_adam_muon_t-1_transformer.jpeg)
New research reveals that different optimization algorithms can produce functionally different financial models with surprisingly similar predictive power.
Can AI Truly Research? A New Benchmark Puts Agents to the Test
A challenging new dataset, DeepResearch-9K, reveals significant limitations in current artificial intelligence systems when it comes to performing complex, multi-step research tasks.
The Illusion of Intelligence: Unmasking Deceptive AI APIs
A new study reveals that many third-party services offering access to powerful language models are riddled with inconsistencies and outright model substitutions, raising serious concerns about reliability and reproducibility.
Teaching AI to Understand Finance: A New Approach to Data Creation
This research details a method for building specialized training data to enhance the reasoning capabilities of artificial intelligence in complex financial scenarios.