It is believed that in knowledge distillation (KD), the role of the teacher is to provide an estimate for the unknown Bayes conditional probability distribution (BCPD) to be used in the student training process. Conventionally, this estimate is obtained by training the teacher using maximum log-likelihood (MLL) method. To improve this estimate for KD, in this paper we introduce the concept of conditional mutual information (CMI) into the estimation of BCPD and propose a novel estimator called the maximum CMI (MCMI) method. Specifically, in MCMI estimation, both the log-likelihood and CMI of the teacher are simultaneously maximized when the teacher is trained. Through Eigen-CAM, it is further shown that maximizing the teacher's CMI value allows the teacher to capture more contextual information in an image cluster. Via conducting a thorough set of experiments, we show that by employing a teacher trained via MCMI estimation rather than one trained via MLL estimation in various state-of-the-art KD frameworks, the student's classification accuracy consistently increases, with the gain of up to 3.32%. This suggests that the teacher's BCPD estimate provided by MCMI method is more accurate than that provided by MLL method. In addition, we show that such improvements in the student's accuracy are more drastic in zero-shot and few-shot settings. Notably, the student's accuracy increases with the gain of up to 5.72% when 5% of the training samples are available to the student (few-shot), and increases from 0% to as high as 84% for an omitted class (zero-shot). Knowledge distillation (Buciluǎ et al., 2006; Hinton et al., 2015) (KD) has received tremendous attention from both academia and industry in recent years as a highly effective model compression technique, and has been deployed in different settings (Radosavovic et al., 2018; Furlanello et al., 2018; Xie et al., 2020). The crux of KD is to distill the knowledge of a cumbersome model (teacher) into a lightweight model (student). One critical component of KD that has received relatively little attention is the training of the teacher model. In fact, in most of the existing KD methods, the teacher is trained to maximize its own performance, even though this does not necessarily lead to an improvement in the student's performance (Cho & Hariharan, 2019; Mirzadeh et al., 2020).

Large language models' (LLMs) abilities are drawn from their pretraining data, and model development begins with data curation. However, decisions around what data is retained or removed during this initial stage is under-scrutinized. In our work, we ground web text, which is a popular pretraining data source, to its social and geographic contexts. We create a new dataset of 10.3 million self-descriptions of website creators, and extract information about who they are and where they are from: their topical interests, social roles, and geographic affiliations. Then, we conduct the first study investigating how ten "quality" and English language identification (langID) filters affect webpages that vary along these social dimensions. Our experiments illuminate a range of implicit preferences in data curation: we show that some quality classifiers act like topical domain filters, and langID can overlook English content from some regions of the world. Overall, we hope that our work will encourage a new line of research on pretraining data curation practices and its social implications.

Tabular data prediction is a fundamental machine learning task for many applications. Existing methods predominantly employ discriminative modeling and operate under the assumption of a fixed target column, necessitating re-training for every new predictive task. Inspired by the generative power of large language models (LLMs), this paper exploits the idea of building universal tabular data predictors based on generative modeling, namely UniPredict. Here, we demonstrate the scalability of an LLM to extensive tabular datasets, enabling it to comprehend diverse tabular inputs and predict target variables following the provided instructions. Specifically, we train a single LLM on an aggregation of 169 tabular datasets with diverse targets and compare its performance against baselines that are trained on each dataset separately. We observe this versatile UniPredict model demonstrates an advantage over other models, ranging from 5.4% to 13.4%, when compared with the best tree-boosting baseline and the best neural network baseline, respectively. We further test UniPredict in few-shot learning settings on another 62 tabular datasets. Our method achieves strong performance in quickly adapting to new tasks. In low-resource few-shot setup, we observed a 100%+ performance advantage compared with XGBoost, and significant margin over all baselines. We envision that UniPredict sheds light on developing a universal tabular data prediction system that learns from data at scale and serves a wide range of prediction tasks.

Recent advances in the performance of large language models (LLMs) have sparked debate over whether, given sufficient training, high-level human abilities emerge in such generic forms of artificial intelligence (AI). Despite the exceptional performance of LLMs on a wide range of tasks involving natural language processing and reasoning, there has been sharp disagreement as to whether their abilities extend to more creative human abilities. A core example is the ability to interpret novel metaphors. Given the enormous and non curated text corpora used to train LLMs, a serious obstacle to designing tests is the requirement of finding novel yet high quality metaphors that are unlikely to have been included in the training data. Here we assessed the ability of GPT4, a state of the art large language model, to provide natural-language interpretations of novel literary metaphors drawn from Serbian poetry and translated into English. Despite exhibiting no signs of having been exposed to these metaphors previously, the AI system consistently produced detailed and incisive interpretations. Human judges, blind to the fact that an AI model was involved, rated metaphor interpretations generated by GPT4 as superior to those provided by a group of college students. In interpreting reversed metaphors, GPT4, as well as humans, exhibited signs of sensitivity to the Gricean cooperative principle. In addition, for several novel English poems GPT4 produced interpretations that were rated as excellent or good by a human literary critic. These results indicate that LLMs such as GPT4 have acquired an emergent ability to interpret complex metaphors, including those embedded in novel poems.

Knowledge distillation (KD) is widely used for compressing a teacher model to reduce its inference cost and memory footprint, by training a smaller student model. However, current KD methods for auto-regressive sequence models suffer from distribution mismatch between output sequences seen during training and those generated by the student during inference. To address this issue, we introduce Generalized Knowledge Distillation (GKD). Instead of solely relying on a fixed set of output sequences, GKD trains the student on its self-generated output sequences by leveraging feedback from the teacher on such sequences. Unlike supervised KD approaches, GKD also offers the flexibility to employ alternative loss functions between the student and teacher, which can be useful when the student lacks the expressivity to mimic the teacher's distribution. Furthermore, GKD facilitates the seamless integration of distillation with RL fine-tuning (RLHF). We demonstrate the efficacy of GKD for distilling auto-regressive language models on summarization, translation, and arithmetic reasoning tasks, and task-agnostic distillation for instruction-tuning.

In this paper, we present an LLM-based code translation method and an associated tool called CoTran, that translates whole-programs from one high-level programming language to another. Current LLM-based code translation methods lack a training approach to ensure that the translated code reliably compiles or bears substantial functional equivalence to the input code. In our work, we train an LLM via reinforcement learning, by modifying the fine-tuning process to incorporate compiler feedback and symbolic execution (symexec)-based equivalence testing feedback that checks for functional equivalence between the input and output programs. The idea is to guide an LLM-in-training, via compiler and symexec-based testing feedback, by letting it know how far it is from producing perfect translations. We report on extensive experiments comparing CoTran with 14 other code translation tools that include human-written transpilers, LLM-based translation tools, and ChatGPT over a benchmark of more than 57,000 Java-Python equivalent pairs, and we show that CoTran outperforms them on relevant metrics such as compilation accuracy (CompAcc) and functional equivalence accuracy (FEqAcc). For example, our tool achieves 48.68% FEqAcc, 76.98% CompAcc for Python-to-Java translation, whereas the nearest competing tool (PLBART-base) only gets 38.26% and 75.77% resp. Also, built upon CodeT5, CoTran achieves +11.23%, +14.89% improvement on FEqAcc and +4.07%, +8.14% on CompAcc for Java-to-Python and Python-to-Java translation resp.

Despite the general capabilities of large pretrained language models, they consistently benefit from further adaptation to better achieve desired behaviors. However, tuning these models has become increasingly resource-intensive, or impossible when model weights are private. We introduce proxy-tuning, a lightweight decoding-time algorithm that operates on top of black-box LMs to achieve the result of directly tuning the model, but by accessing only its prediction over the output vocabulary. Our method instead tunes a smaller LM, then applies the difference between the predictions of the small tuned and untuned LMs to shift the original predictions of the base model in the direction of tuning, while retaining the benefits of larger scale pretraining. In experiments, when we apply proxy-tuning to Llama2-70B using proxies of only 7B size, we can close 88% of the gap between Llama2-70B and its truly-tuned chat version, when evaluated across knowledge, reasoning, and safety benchmarks. Interestingly, when tested on TruthfulQA, proxy-tuned models are actually more truthful than directly tuned models, possibly because decoding-time guidance better retains the model's factual knowledge. We then demonstrate the generality of proxy-tuning by applying it for domain adaptation on code, and task-specific finetuning on question-answering and math problems. Our work demonstrates the promise of using small tuned LMs to efficiently customize large, potentially proprietary LMs through decoding-time guidance.

The output tendencies of Pre-trained Language Models (PLM) vary markedly before and after Fine-Tuning (FT) due to the updates to the model parameters. These divergences in output tendencies result in a gap in the social biases of PLMs. For example, there exits a low correlation between intrinsic bias scores of a PLM and its extrinsic bias scores under FT-based debiasing methods. Additionally, applying FT-based debiasing methods to a PLM leads to a decline in performance in downstream tasks. On the other hand, PLMs trained on large datasets can learn without parameter updates via In-Context Learning (ICL) using prompts. ICL induces smaller changes to PLMs compared to FT-based debiasing methods. Therefore, we hypothesize that the gap observed in pre-trained and FT models does not hold true for debiasing methods that use ICL. In this study, we demonstrate that ICL-based debiasing methods show a higher correlation between intrinsic and extrinsic bias scores compared to FT-based methods. Moreover, the performance degradation due to debiasing is also lower in the ICL case compared to that in the FT case.

This study explores the learning dynamics of neural networks by analyzing the singular value decomposition (SVD) of their weights throughout training. Our investigation reveals that an orthogonal basis within each multidimensional weight's SVD representation stabilizes during training. Building upon this, we introduce Orthogonality-Informed Adaptive Low-Rank (OIALR) training, a novel training method exploiting the intrinsic orthogonality of neural networks. OIALR seamlessly integrates into existing training workflows with minimal accuracy loss, as demonstrated by benchmarking on various datasets and well-established network architectures. With appropriate hyperparameter tuning, OIALR can surpass conventional training setups, including those of state-of-the-art models.

We propose a novel, heterogeneous multi-agent architecture that miniaturizes rovers by outsourcing power generation to a central hub. By delegating power generation and distribution functions to this hub, the size, weight, power, and cost (SWAP-C) per rover are reduced, enabling efficient fleet scaling. As these rovers conduct mission tasks around the terrain, the hub charges an array of replacement battery modules. When a rover requires charging, it returns to the hub to initiate an autonomous docking sequence and exits with a fully charged battery. This confers an advantage over direct charging methods, such as wireless or wired charging, by replenishing a rover in minutes as opposed to hours, increasing net rover uptime. This work shares an open-source platform developed to demonstrate battery swapping on unknown field terrain. We detail our design methodologies utilized for increasing system reliability, with a focus on optimization, robust mechanical design, and verification. Optimization of the system is discussed, including the design of passive guide rails through simulation-based optimization methods which increase the valid docking configuration space by 258%. The full system was evaluated during integrated testing, where an average servicing time of 98 seconds was achieved on surfaces with a gradient up to 10{\deg}. We conclude by briefly proposing flight considerations for advancing the system toward a space-ready design. In sum, this prototype represents a proof of concept for autonomous docking and battery transfer on field terrain, advancing its Technology Readiness Level (TRL) from 1 to 3.