Hyponatremia: Predict whether a hyponatremia lab comes back as normal (>=135 mmol/L), mild (>=130 and <135 mmol/L), moderate (>=125 and <130 mmol/L), or severe (<125 mmol/L). We consider all lab results coded as LOINC/LG11363-5, LOINC/2951-2, or LOINC/2947-0. Anemia: Predict whether an anemia lab comes back as normal (>=120 g/L), mild (>=110 and <120 g/L), moderate (>=70 and <110 g/L), or severe (<70 g/L). We consider all lab results coded as LOINC/LP392452-1. Please note that for the results of our baseline experiments in Section 5, we reframe these lab value tasks as binary classification tasks, where a label is "negative" if the result is normal and "positive" otherwise.

From CNNs toattention mechanisms, encoding inductivebiases intoneural networks has been a fruitful source of improvement in machine learning.

However,inmanyreal-worldtasks,the learned function is intended to satisfy domain-specific constraints. We focus on monotonicity constraints, which are common and require that the function's output increases with increasing values of specific input features.

From CNNs to attention mechanisms, encoding inductive biases into neural networks has been a fruitful source of improvement in machine learning. Adding auxiliary losses to the main objective function is a general way of encoding biases that can help networks learn better representations. However, since auxiliary losses are minimized only on training data, they suffer from the same generalization gap as regular task losses. Moreover, by adding a term to the loss function, the model optimizes a different objective than the one we care about. In this work we address both problems: first, we take inspiration from transductive learning and note that after receiving an input but before making a prediction, we can fine-tune our networks on any unsupervised loss. We call this process tailoring, because we customize the model to each input to ensure our prediction satisfies the inductive bias. Second, we formulate meta-tailoring, a nested optimization similar to that in meta-learning, and train our models to perform well on the task objective after adapting them using an unsupervised loss. The advantages of tailoring and meta-tailoring are discussed theoretically and demonstrated empirically on a diverse set of examples.

Assembly-to-source code translation is a critical task in reverse engineering, cybersecurity, and software maintenance, yet systematic benchmarks for evaluating large language models on this problem remain scarce. In this work, we present the first comprehensive evaluation of five state-of-the-art large language models on assembly-to-source translation. We assess model performance using a diverse set of metrics capturing lexical similarity (BLEU, ROUGE, and METEOR), semantic alignment (BERTScore), fluency (Perplexity), and efficiency (time prediction). Our results reveal clear trade-offs: while certain models excel in text similarity metrics, others demonstrate lower perplexity or faster inference times. We further provide qualitative analyses of typical model successes and failure cases, highlighting challenges such as control flow recovery and identifier reconstruction. Taken together, our benchmark offers actionable insights into the strengths and limitations of current large language models for program translation, establishing a foundation for future research in combining accuracy with efficiency for real-world applications.

For [23] we refer to the paper pointed by you. 5. To Reviewer_25: the overall conceptual motivation for the paper is somewhat weak... Nystrom approximation can be used to approximate the kernel matrix and speed up kernel machines, and from Table 1 we can see that the performance is suboptimal even when rank=200 (see the 5-th column). In this case, it requires 200 inner product computations to make one prediction, which is too slow for many real-time systems (e.g., web applications, robotic applications ...). Therefore state-of-the-art Nystrom method is not good enough, and we reduce the prediction time to 10~20 inner products with a better classification accuracy, which is a big improvement. Also, as we mentioned in the point 1 above, although we want to optimize the prediction time, our method still has fast training time. We agree that the psuedo landmark point technique can be potentially applied to speed up the training time, and it is an interesting research direction.