Current Large Language Models (LLMs) are usually post-trained on large-scale carefully curated datasets (data & training scaling) and doing reasoning in test time (inference time scaling). In this work, we present a new scaling paradigm, Input-Time Scaling, to complement previous scaling methods by putting resources on queries (input time). During training and testing, we utilize meta-knowledge from LLMs to refine inputs with different strategies. We also discover a new phenomenon, train-test co-design. It requires us to apply query strategies during training and testing as a whole. Only applying strategies on training or testing would seriously degrade the performance gained. We are also surprised to find that seemingly low data quality datasets can perform better. We can get the best performance even by adding irrelevant information to the queries, with randomly selected 1k examples from a minimally filtered dataset. These findings contradict the widely held inductive bias, "garbage in, garbage out". Curating datasets with seemingly high-quality data can even potentially limit the performance ceiling. In addition, models trained on more data with similar quality (15k VS 1k) perform worse, the intuition of simply scaling the size should also be carefully inspected. The good news is that our findings are compatible with the Less is More phenomenon. 1K examples are enough to invoke high-level reasoning ability. With experiments on Qwen2.5-32B-Instruct, we are able to reach SOTA performance among 32B models on AIME24(76.7%) and AIME25(76.7%) pass@1. We can further achieve AIME24(76.7%) and AIME25(80%) with a majority vote of three models. Starting from DeepSeek-R1-Distill-Qwen-32B, the result would be 90.0% on AIME24 and 80.0% on AIME25. To facilitate reproducibility and further research, we are working on open-source our datasets, data pipelines, evaluation results, and checkpoints.

We investigated the use of Empirical Mode Decomposition (EMD) combined with Gaussian Mixture Models (GMM), feature engineering and machine learning algorithms to optimize trading decisions. We used five, two, and one year samples of hourly candle data for GameStop, Tesla, and XRP (Ripple) markets respectively. Applying a 15 hour rolling window for each market, we collected several features based on a linear model and other classical features to predict the next hour's movement. Subsequently, a GMM filtering approach was used to identify clusters among these markets. For each cluster, we applied the EMD algorithm to extract high, medium, low and trend components from each feature collected. A simple thresholding algorithm was applied to classify market movements based on the percentage change in each market's close price. We then evaluated the performance of various machine learning models, including Random Forests (RF) and XGBoost, in classifying market movements. A naive random selection of trading decisions was used as a benchmark, which assumed equal probabilities for each outcome, and a temporal cross-validation approach was used to test models on 40%, 30%, and 20% of the dataset. Our results indicate that transforming selected features using EMD improves performance, particularly for ensemble learning algorithms like Random Forest and XGBoost, as measured by accumulated profit. Finally, GMM filtering expanded the range of learning algorithm and data source combinations that outperformed the top percentile of the random baseline.

The decisions traders make to buy or sell an asset depend on various analyses, with expertise required to identify patterns that can be exploited for profit. In this paper we identify novel features extracted from emergent and well-established financial markets using linear models and Gaussian Mixture Models (GMM) with the aim of finding profitable opportunities. We used approximately six months of data consisting of minute candles from the Bitcoin, Pepecoin, and Nasdaq markets to derive and compare the proposed novel features with commonly used ones. These features were extracted based on the previous 59 minutes for each market and used to identify predictions for the hour ahead. We explored the performance of various machine learning strategies, such as Random Forests (RF) and K-Nearest Neighbours (KNN) to classify market movements. A naive random approach to selecting trading decisions was used as a benchmark, with outcomes assumed to be equally likely. We used a temporal cross-validation approach using test sets of 40%, 30% and 20% of total hours to evaluate the learning algorithms' performances. Our results showed that filtering the time series facilitates algorithms' generalisation. The GMM filtering approach revealed that the KNN and RF algorithms produced higher average returns than the random algorithm.

Auditory attention decoding (AAD) aims to extract from brain activity the attended speaker amidst candidate speakers, offering promising applications for neuro-steered hearing devices and brain-computer interfacing. This pilot study makes a first step towards AAD using the noise-tagging stimulus protocol, which evokes reliable code-modulated evoked potentials, but is minimally explored in the auditory modality. Participants were sequentially presented with two Dutch speech stimuli that were amplitude-modulated with a unique binary pseudo-random noise-code, effectively tagging these with additional decodable information. We compared the decoding of unmodulated audio against audio modulated with various modulation depths, and a conventional AAD method against a standard method to decode noise-codes. Our pilot study revealed higher performances for the conventional method with 70 to 100 percent modulation depths compared to unmodulated audio. The noise-code decoder did not further improve these results. These fundamental insights highlight the potential of integrating noise-codes in speech to enhance auditory speaker detection when multiple speakers are presented simultaneously.

Intel's Core i9-14900K still offers some of the best performance around -- albeit at a similarly beastly power draw -- but offers negligible performance improvement over its direct predecessor, the 13900K. New support for Wi-Fi 7, Thunderbolt 5, and performance-boosting AI features are a nice touch, though. A new generation of refreshed Raptor Lake processors have arrived. After months of rumors and leaks--and an official announcement just yesterday--Intel's latest batch of desktop CPUs take their place as the 14th generation in the Core lineup. You can catch up on the specs and speeds in our comprehensive coverage of the unveiling, but the basics are straightforward. Six new chips have launched, with two variants each of unlocked Core i9, Core i7, and Core i5 parts.

Prompt tuning is an effective way to adapt the pre-trained visual-language model (VLM) to the downstream task using task-related textual tokens. Representative CoOp-based work combines the learnable textual tokens with the class tokens to obtain specific textual knowledge. However, the specific textual knowledge is the worse generalization to the unseen classes because it forgets the essential general textual knowledge having a strong generalization ability. To tackle this issue, we introduce a novel Knowledge-guided Context Optimization (KgCoOp) to enhance the generalization ability of the learnable prompt for unseen classes. The key insight of KgCoOp is that forgetting about essential knowledge can be alleviated by reducing the discrepancy between the learnable prompt and the hand-crafted prompt. Especially, KgCoOp minimizes the discrepancy between the textual embeddings generated by learned prompts and the hand-crafted prompts. Finally, adding the KgCoOp upon the contrastive loss can make a discriminative prompt for both seen and unseen tasks. Extensive evaluation of several benchmarks demonstrates that the proposed Knowledge-guided Context Optimization is an efficient method for prompt tuning, \emph{i.e.,} achieves better performance with less training time.

NVIDIA A100 HBM Ampere GPU 80GB premiere the world's fastest memory bandwidth at over 2 terabytes per second to run the largest simulation models and datasets. It allows researchers to quickly deliver accurate results and deploy solutions into production at scale. NVIDIA A100 Tensor Cores with Tensor Float (TF32) provide up to 20x higher performance over the NVIDIA Volta with zero code changes and an additional 2x boost with automatic mixed precision and FP16. For the largest models with enormous data tables like deep learning recommendation models (DLRM), Ampere A100 80GB GPU reaches 1.3 TB of unified memory per node and delivers up to a 3x throughput increase over A100 40GB GPU. In MLPerf, it has set multiple performance records in the industry-wide benchmark for AI training.

The goal of chip design always has been to optimize power, performance, and area (PPA), but results can vary greatly even with the best tools and highly experienced engineering teams. Optimizing PPA involves a growing number of tradeoffs that can vary by application, by availability of IP and other components, as well as the familiarity of engineers with different tools and methodologies. For example, higher performance may be achieved with a larger processor, but it also can be done using smaller, more specialized processing elements with tighter integration of hardware and software. So even in the same area and with the same power budget, there are different ways of achieving the same goal, and the optimum mix may vary depending upon a specific domain or vendor's needs. This is made even more complex by the rising demand for security.

Start-up AlphaIC yesterday announced that it has begun sampling its Gluon coprocessor for edge AI inference to customers. AlphaIC claims the chip delivers competitive performance compared to Nvidia in vision workloads such as object detection. Gluon is based on AlphaIC's proprietary architecture that has an instruction set architecture (ISA) optimized for AI, and has been in development for two years. The ISA refers to the instructions that a chip can process. AlphaIC calls it Real AI Processor (RAT) architecture.