We introduce Bootstrap Your Own Latent (BYOL), a new approach to self-supervised image representation learning. BYOL relies on two neural networks, referred to as online and target networks, that interact and learn from each other. From an augmented view of an image, we train the online network to predict the target network representation of the same image under a different augmented view. At the same time, we update the target network with a slow-moving average of the online network. While state-of-the art methods intrinsically rely on negative pairs, BYOL achieves a new state of the art without them. BYOL reaches 74.3% top-1 classification accuracy on ImageNet using the standard linear evaluation protocol with a standard ResNet-50 architecture and 79.6% with a larger ResNet. We also show that BYOL performs on par or better than the current state of the art on both transfer and semi-supervised benchmarks.

Continual learning agents experience a stream of (related) tasks. The main challenge is that the agent must not forget previous tasks and also adapt to novel tasks in the stream. We are interested in the intersection of two recent continual-learning scenarios. In meta-continual learning, the model is pre-trained using meta-learning to minimize catastrophic forgetting of previous tasks. In continual-meta learning, the aim is to train agents for faster remembering of previous tasks through adaptation.

Reward shaping is an effective technique for incorporating domain knowledge into reinforcement learning (RL). Existing approaches such as potential-based reward shaping normally make full use of a given shaping reward function. However, since the transformation of human knowledge into numeric reward values is often imperfect due to reasons such as human cognitive bias, completely utilizing the shaping reward function may fail to improve the performance of RL algorithms. In this paper, we consider the problem of adaptively utilizing a given shaping reward function. We formulate the utilization of shaping rewards as a bi-level optimization problem, where the lower level is to optimize policy using the shaping rewards and the upper level is to optimize a parameterized shaping weight function for true reward maximization. We formally derive the gradient of the expected true reward with respect to the shaping weight function parameters and accordingly propose three learning algorithms based on different assumptions. Experiments in sparse-reward cartpole and MuJoCo environments show that our algorithms can fully exploit beneficial shaping rewards, and meanwhile ignore unbeneficial shaping rewards or even transform them into beneficial ones.

The challenge of separating AI-generated text from human-authored content is becoming more urgent as generative AI technologies like ChatGPT become more widely available. In this work, we address this issue by looking at both the detection of content that has been entirely generated by AI and the identification of human text that has been reworded by AI. In our work, a comprehensive methodology to detect AI- generated text using XLM-RoBERTa, a state-of-the-art multilingual transformer model. Our approach includes rigorous preprocessing, and feature extraction involving perplexity, semantic, and readability features. We fine-tuned the XLM-RoBERTa model on a balanced dataset of human and AI-generated texts and evaluated its performance. The model demonstrated high accuracy and robust performance across various text genres. Additionally, we conducted feature analysis to understand the model's decision-making process, revealing that perplexity and attention-based features are critical in differentiating between human and AI-generated texts. Our findings offer a valuable tool for maintaining academic integrity and contribute to the broader field of AI ethics by promoting transparency and accountability in AI systems. Future research directions include exploring other advanced models and expanding the dataset to enhance the model's generalizability.

A startup hopes to challenge Nvidia, AMD, and Intel with a chip that wrangles probabilities rather than ones and zeros. Extropic claims its exotic new chip, called XTR-0, could be thousands of times more energy efficient than existing chips when scaled up. Extropic, a startup developing an exotic new kind of computer chip that handles probabilistic bits, has produced its first working hardware along with proof that more advanced systems will tackle useful tasks in artificial intelligence and scientific research. The startup's chips work in a fundamentally different way to chips from Nvidia, AMD, and others, and promise to be thousands of times more energy efficient when scaled up. With AI companies pouring billions of dollars into building datacenters, a completely new approach could offer a far less costly alternative to vast arrays of conventional chips.

In Indiana, Exploration Acres found a way to keep the family farm alive. Exploration Acres has operated its award-winning corn maze for almost 20 years. Breakthroughs, discoveries, and DIY tips sent every weekday. The adage refers to a farmer's goal for their crops if they hope to make the October harvest. And while most Midwesterners are familiar with the axiom, Tim Fitzgerald knows the folksy refrain lost its relevancy decades ago.

Is any special feature operation applied in ETN? & Does a larger K help? The motivation to compute affinity matrices & How to achieve the error diffusion. Please see Figure 1 in submission for example. Performance issues, including increased training burden and running time. Thanks for pointing out the mistake in real-time stylization, which will be corrected in revision.

Test-time scaling offers a promising path to improve LLM reasoning by utilizing more compute at inference time; however, the true promise of this paradigm lies in extrapolation (i.e., improvement in performance on hard problems as LLMs keep "thinking" for longer, beyond the maximum token budget they were trained on). Surprisingly, we find that most existing reasoning models do not extrapolate well. We show that one way to enable extrapolation is by training the LLM to perform in-context exploration: training the LLM to effectively spend its test time budget by chaining operations (such as generation, verification, refinement, etc.), or testing multiple hypotheses before it commits to an answer. To enable in-context exploration, we identify three key ingredients as part of our recipe e3: (1) chaining skills that the base LLM has asymmetric competence in, e.g., chaining verification (easy) with generation (hard), as a way to implement in-context search; (2) leveraging "negative" gradients from incorrect traces to amplify exploration during RL, resulting in longer search traces that chains additional asymmetries; and (3) coupling task difficulty with training token budget during training via a specifically-designed curriculum to structure in-context exploration. Our recipe e3 produces the best known 1.7B model according to AIME'25 and HMMT'25 scores, and extrapolates to 2x the training token budget. Our e3-1.7B model not only attains high pass@1 scores, but also improves pass@k over the base model.

Weaknesses: My main concern with the submission is that the evaluation scenario OSAKA seems too specific and designed primarily for a set of algorithms in between Meta- & Continual-Learning while failing to make a broader argument for other approaches to Continual Learning. While certain aspects of OSAKA are certainly desirable (OOD tasks, Unknown task changes, Online Evaluation) there is a strong assumption made in allowing for Pre-training that may not be adequate in certain CL settings, limiting the generality of OSAKA. Furthermore, it is unclear how aspects such as controllable non-stationarity would be implemented in Reinforcement Learning. Furthermore, I personally feel that if task-revisiting is to be implemented, new OOD tasks should be designed in a way that explicitly re-uses skills that can be learned on a previous problem in a novel setting, instead of merely re-visiting the problem without modification. The problem with this assumption in general is that Catastrophic Forgetting is significantly reduced through an implicit form of replay provided by the environment, making it difficult to tell to which extent catastrophic forgetting is actually a problem of these algorithms.

This review paper explores recent advancements and emerging approaches in Information Retrieval (IR) applied to Natural Language Processing (NLP). We examine traditional IR models such as Boolean, vector space, probabilistic, and inference network models, and highlight modern techniques including deep learning, reinforcement learning, and pretrained transformer models like BERT. We discuss key tools and libraries - Lucene, Anserini, and Pyserini - for efficient text indexing and search. A comparative analysis of sparse, dense, and hybrid retrieval methods is presented, along with applications in web search engines, cross-language IR, argument mining, private information retrieval, and hate speech detection. Finally, we identify open challenges and future research directions to enhance retrieval accuracy, scalability, and ethical considerations.