Learning visual representations from natural language supervision has recently shown great promise in a number of pioneering works. In general, these language-augmented visual models demonstrate strong transferability to a variety of datasets and tasks. However, it remains challenging to evaluate the transferablity of these models due to the lack of easy-to-use evaluation toolkits and public benchmarks. To tackle this, we buildELEVATER 1, the first benchmark and toolkit for evaluating (pre-trained) language-augmented visual models. ELEVATERis composed of three components.

Knowledge distillation is an effective approach to learn compact models (students) with the supervision of large and strong models (teachers). As empirically there exists a strong correlation between the performance of teacher and student models, it is commonly believed that a high performing teacher is preferred. Consequently, practitioners tend to use a well trained network or an ensemble of them as the teacher. In this paper, we observe that an intermediate model, i.e., a checkpoint in the middle of the training procedure, often serves as a better teacher compared to the fully converged model, although the former has much lower accuracy. More surprisingly, a weak snapshot ensemble of several intermediate models from a same training trajectory can outperform a strong ensemble of independently trained and fully converged models, when they are used as teachers. We show that this phenomenon can be partially explained by the information bottleneck principle: the feature representations of intermediate models can have higher mutual information regarding the input, and thus contain more ``dark knowledge'' for effective distillation. We further propose an optimal intermediate teacher selection algorithm based on maximizing the total task-related mutual information. Experiments verify its effectiveness and applicability.

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.

Machine learning in neurosurgery is limited by challenges in assembling large, high-quality imaging datasets. Synthetic data offers a scalable, privacy-preserving solution. We evaluated the feasibility of generating realistic lateral cervical spine radiographs using a denoising diffusion probabilistic model (DDPM) trained on 4,963 images from the Cervical Spine X-ray Atlas. Model performance was monitored via training/validation loss and Frechet inception distance, and synthetic image quality was assessed in a blinded "clinical Turing test" with six neuroradiologists and two spine-fellowship trained neurosurgeons. Experts reviewed 50 quartets containing one real and three synthetic images, identifying the real image and rating realism on a 4-point Likert scale. Experts correctly identified the real image in 29% of trials (Fleiss' kappa=0.061). Mean realism scores were comparable between real (3.323) and synthetic images (3.228, 3.258, and 3.320; p=0.383, 0.471, 1.000). Nearest-neighbor analysis found no evidence of memorization. We also provide a dataset of 20,063 synthetic radiographs. These results demonstrate that DDPM-generated cervical spine X-rays are statistically indistinguishable in realism and quality from real clinical images, offering a novel approach to creating large-scale neuroimaging datasets for ML applications in landmarking, segmentation, and classification.

We curate a 16 - class corpus (~47k images) by merging Kaggle, ImageNet, and web - cr awled data, and create six balanced variants via SMOTE oversampling and targeted undersampling. Lightweight ensembles, such as Random Forest, AdaBoost, and a soft - voting combiner built on MobileNet - V2 features are benchmarked against a configurable ResNet - style CNN trained with strong augmentation and label smoothing. The best ensemble (SMOTE - combined) attains 74.8% test accuracy, while the CNN achieves 79.19% on the full test set and 81.25% on an unseen inferen ce batch, confirming the advantage of deep models. Nonetheless, the most under - represented class (Barge) remains a failure mode, highlighting the limits of rebalancing alone. Results suggest prioritizing additional minority - class collection and cost - sensit ive objectives (e.g., focal loss) and exploring hybrid ensemble or CNN pipelines to combine interpretability with representational power. The best ensemble (SMOTE - combined) reached 74.8% test accuracy, while the final checkpoint of CNN achieved 79.1 9 % on the full test set and 81. 25 % on an unseen EE531 inference batch, confirming that deep models excel overall but still falter on the most under - represented class ( Barge), underscoring the persistent challenge of extreme imbalance.

This project performs multimodal sentiment analysis using the CMU-MOSEI dataset, using transformer-based models with early fusion to integrate text, audio, and visual modalities. We employ BERTbased encoders for each modality, extracting embed-dings that are concatenated before classification. The model achieves strong performance, with 97.87% 7-class accuracy and a 0.9682 F1-score on the test set, demonstrating the effectiveness of early fusion in capturing cross-modal interactions. The training utilized Adam optimization (lr=1e-4), dropout (0.3), and early stopping to ensure generalization and robustness. Results highlight the superiority of transformer architectures in modeling multimodal sentiment, with a low MAE (0.1060) indicating precise sentiment intensity prediction. Future work may compare fusion strategies or enhance interpretability.

Modern language models often have open weights but closed training data. We formalize the problem of data approximation from model weights and propose several baselines and metrics. We develop a gradient-based approach that selects the highest-matching data from a large public text corpus and show its effectiveness at recovering useful data given only weights of the original and finetuned models. Even when none of the true training data is known, our method is able to locate a small subset of public Web documents can be used to train a model to close to the original model performance given models trained for both classification and supervised-finetuning. On the AG News classification task, our method improves performance from 65% (using randomly selected data) to 80%, approaching the expert benchmark of 88%. When applied to a model trained with SFT on MSMARCO web documents, our method reduces perplexity from 3.3 to 2.3, compared to an expert LLAMA model's perplexity of 2.0.

A fundamental open challenge in modern LLM scaling is the lack of understanding around emergent capabilities. In particular, language model pretraining loss is known to be highly predictable as a function of compute. However, downstream capabilities are far less predictable -- sometimes even exhibiting emergent jumps -- which makes it challenging to anticipate the capabilities of future models. In this work, we first pose the task of emergence prediction: given access to current LLMs that have random few-shot accuracy on a task, can we predict whether future models (GPT-N+1) will have non-trivial accuracy on that task? We then discover a simple insight for this problem: finetuning LLMs on a given task can shift the point in scaling at which emergence occurs towards less capable models. To operationalize this insight, we can finetune LLMs with varying amounts of data and fit a parametric function that predicts when emergence will occur (i.e., "emergence laws"). We validate this approach using four standard NLP benchmarks where large-scale open-source LLMs already demonstrate emergence (MMLU, GSM8K, CommonsenseQA, and CoLA). Using only small-scale LLMs, we find that, in some cases, we can accurately predict whether models trained with up to 4x more compute have emerged. Finally, we present a case study of two realistic uses for emergence prediction.

Therefore, directly sampling from the models, comparing the methods or becoming acquainted with them may present challenges. To mitigate this issue we introduce models - contains the models currently implemented Symbotunes, an open-source unified hub for symbolic in our hub. Each model is in a separate sub-directory, music generative models. Symbotunes contains modern which also contains an example training script, Python implementations of well-known methods for data - contains all the data handling utilities available symbolic music generation, as well as a unified pipeline in the hub - datasets, tokenizers, and data transforms.

Knowledge distillation is an effective approach to learn compact models (students) with the supervision of large and strong models (teachers). As empirically there exists a strong correlation between the performance of teacher and student models, it is commonly believed that a high performing teacher is preferred. Consequently, practitioners tend to use a well trained network or an ensemble of them as the teacher. In this paper, we observe that an intermediate model, i.e., a checkpoint in the middle of the training procedure, often serves as a better teacher compared to the fully converged model, although the former has much lower accuracy. More surprisingly, a weak snapshot ensemble of several intermediate models from a same training trajectory can outperform a strong ensemble of independently trained and fully converged models, when they are used as teachers.