Search

Prior work has shown the existence of contextual neurons in language models, including a neuron that activates on German text. We show that this neuron exists within a broader contextual n-gram circuit: we find late layer neurons which recognize and continue n-grams common in German text, but which only activate if the German neuron is active. We investigate the formation of this circuit throughout training and find that it is an example of what we call a second-order circuit. In particular, both the constituent n-gram circuits and the German detection circuit which culminates in the German neuron form with independent functions early in training - the German detection circuit partially through modeling German unigram statistics, and the n-grams by boosting appropriate completions. Only after both circuits have already formed do they fit together into a second-order circuit. Contrary to the hypotheses presented in prior work, we find that the contextual n-gram circuit forms gradually rather than in a sudden phase transition. We further present a range of anomalous observations such as a simultaneous phase transition in many tasks coinciding with the learning rate warm-up, and evidence that many context neurons form simultaneously early in training but are later unlearned., Comment: Accepted workshop paper at ATTRIB 2023 (@ NeurIPS)

We propose a framework for optimizing a planar parallel-jaw gripper for use with multiple objects. While optimizing general-purpose grippers and contact locations for grasps are both well studied, co-optimizing grasps and the gripper geometry to execute them receives less attention. As such, our framework synthesizes grippers optimized to stably grasp sets of polygonal objects. Given a fixed number of contacts and their assignments to object faces and gripper jaws, our framework optimizes contact locations along these faces, gripper pose for each grasp, and gripper shape. Our key insights are to pose shape and contact constraints in frames fixed to the gripper jaws, and to leverage the linearity of constraints in our grasp stability and gripper shape models via an augmented Lagrangian formulation. Together, these enable a tractable nonlinear program implementation. We apply our method to several examples. The first illustrative problem shows the discovery of a geometrically simple solution where possible. In another, space is constrained, forcing multiple objects to be contacted by the same features as each other. Finally a toolset-grasping example shows that our framework applies to complex, real-world objects. We provide a physical experiment of the toolset grasps., Comment: 2023 IEEE IROS conference

Although pre-trained language models encode generic knowledge beneficial for planning and control, they may fail to generate appropriate control policies for domain-specific tasks. Existing fine-tuning methods use human feedback to address this limitation, however, sourcing human feedback is labor intensive and costly. We present a fully automated approach to fine-tune pre-trained language models for applications in autonomous systems, bridging the gap between generic knowledge and domain-specific requirements while reducing cost. The method synthesizes automaton-based controllers from pre-trained models guided by natural language task descriptions. These controllers are verifiable against independently provided specifications within a world model, which can be abstract or obtained from a high-fidelity simulator. Controllers with high compliance with the desired specifications receive higher ranks, guiding the iterative fine-tuning process. We provide quantitative evidences, primarily in autonomous driving, to demonstrate the method's effectiveness across multiple tasks. The results indicate an improvement in percentage of specifications satisfied by the controller from 60% to 90%.

Sentiment is a pervasive feature in natural language text, yet it is an open question how sentiment is represented within Large Language Models (LLMs). In this study, we reveal that across a range of models, sentiment is represented linearly: a single direction in activation space mostly captures the feature across a range of tasks with one extreme for positive and the other for negative. Through causal interventions, we isolate this direction and show it is causally relevant in both toy tasks and real world datasets such as Stanford Sentiment Treebank. Through this case study we model a thorough investigation of what a single direction means on a broad data distribution. We further uncover the mechanisms that involve this direction, highlighting the roles of a small subset of attention heads and neurons. Finally, we discover a phenomenon which we term the summarization motif: sentiment is not solely represented on emotionally charged words, but is additionally summarized at intermediate positions without inherent sentiment, such as punctuation and names. We show that in Stanford Sentiment Treebank zero-shot classification, 76% of above-chance classification accuracy is lost when ablating the sentiment direction, nearly half of which (36%) is due to ablating the summarized sentiment direction exclusively at comma positions.

Recent work has shown that Large Language Models (LLMs) can unintentionally leak sensitive information present in their training data. In this paper, we present Model Perturbations (MoPe), a new method to identify with high confidence if a given text is in the training data of a pre-trained language model, given white-box access to the models parameters. MoPe adds noise to the model in parameter space and measures the drop in log-likelihood at a given point $x$, a statistic we show approximates the trace of the Hessian matrix with respect to model parameters. Across language models ranging from $70$M to $12$B parameters, we show that MoPe is more effective than existing loss-based attacks and recently proposed perturbation-based methods. We also examine the role of training point order and model size in attack success, and empirically demonstrate that MoPe accurately approximate the trace of the Hessian in practice. Our results show that the loss of a point alone is insufficient to determine extractability -- there are training points we can recover using our method that have average loss. This casts some doubt on prior works that use the loss of a point as evidence of memorization or unlearning.

Automatic Program translation has enormous application value and hence has been attracting significant interest from AI researchers. However, we observe that current program translation models still make elementary syntax errors, particularly, when the target language does not have syntax elements in the source language. Metrics like BLUE, CodeBLUE and computation accuracy may not expose these issues. In this paper we introduce a new metrics for programming language translation and these metrics address these basic syntax errors. We develop a novel active defects probing suite called Syntactic Unit Tests (SUT) which includes a highly interpretable evaluation harness for accuracy and test scoring. Experiments have shown that even powerful models like ChatGPT still make mistakes on these basic unit tests. Specifically, compared to previous program translation task evaluation dataset, its pass rate on our unit tests has decreased by 26.15%. Further our evaluation harness reveal syntactic element errors in which these models exhibit deficiencies.

Since their inception Generative Adversarial Networks (GANs) have been popular generative models across images, audio, video, and tabular data. In this paper we study whether given access to a trained GAN, as well as fresh samples from the underlying distribution, if it is possible for an attacker to efficiently identify if a given point is a member of the GAN's training data. This is of interest for both reasons related to copyright, where a user may want to determine if their copyrighted data has been used to train a GAN, and in the study of data privacy, where the ability to detect training set membership is known as a membership inference attack. Unlike the majority of prior work this paper investigates the privacy implications of using GANs in black-box settings, where the attack only has access to samples from the generator, rather than access to the discriminator as well. We introduce a suite of membership inference attacks against GANs in the black-box setting and evaluate our attacks on image GANs trained on the CIFAR10 dataset and tabular GANs trained on genomic data. Our most successful attack, called The Detector, involve training a second network to score samples based on their likelihood of being generated by the GAN, as opposed to a fresh sample from the distribution. We prove under a simple model of the generator that the detector is an approximately optimal membership inference attack. Across a wide range of tabular and image datasets, attacks, and GAN architectures, we find that adversaries can orchestrate non-trivial privacy attacks when provided with access to samples from the generator. At the same time, the attack success achievable against GANs still appears to be lower compared to other generative and discriminative models; this leaves the intriguing open question of whether GANs are in fact more private, or if it is a matter of developing stronger attacks.

Software version migration and program translation are an important and costly part of the lifecycle of large codebases. Traditional machine translation relies on parallel corpora for supervised translation, which is not feasible for program translation due to a dearth of aligned data. Recent unsupervised neural machine translation techniques have overcome data limitations by included techniques such as back translation and low level compiler intermediate representations (IR). These methods face significant challenges due to the noise in code snippet alignment and the diversity of IRs respectively. In this paper we propose a novel model called Code Distillation (CoDist) whereby we capture the semantic and structural equivalence of code in a language agnostic intermediate representation. Distilled code serves as a translation pivot for any programming language, leading by construction to parallel corpora which scale to all available source code by simply applying the distillation compiler. We demonstrate that our approach achieves state-of-the-art performance on CodeXGLUE and TransCoder GeeksForGeeks translation benchmarks, with an average absolute increase of 12.7% on the TransCoder GeeksforGeeks translation benchmark compare to TransCoder-ST.

Machine unlearning, the study of efficiently removing the impact of specific training instances on a model, has garnered increased attention in recent years due to regulatory guidelines such as the \emph{Right to be Forgotten}. Achieving precise unlearning typically involves fully retraining the model and is computationally infeasible in case of very large models such as Large Language Models (LLMs). To this end, recent work has proposed several algorithms which approximate the removal of training data without retraining the model. These algorithms crucially rely on access to the model parameters in order to update them, an assumption that may not hold in practice due to computational constraints or having only query access to the LLMs. In this work, we propose a new class of unlearning methods for LLMs called ``In-Context Unlearning.'' This method unlearns instances from the model by simply providing specific kinds of inputs in context, without the need to update model parameters. To unlearn specific training instances, we present these instances to the LLMs at inference time along with labels that differ from their ground truth. Our experimental results demonstrate that in-context unlearning performs on par with, or in some cases outperforms other state-of-the-art methods that require access to model parameters, effectively removing the influence of specific instances on the model while preserving test accuracy., Comment: Accepted at ICML 2024

We show that language model finetuning can be improved, sometimes dramatically, with a simple augmentation. NEFTune adds noise to the embedding vectors during training. Standard finetuning of LLaMA-2-7B using Alpaca achieves 29.79% on AlpacaEval, which rises to 64.69% using noisy embeddings. NEFTune also improves over strong baselines on modern instruction datasets. Models trained with Evol-Instruct see a 10% improvement, with ShareGPT an 8% improvement, and with OpenPlatypus an 8% improvement. Even powerful models further refined with RLHF such as LLaMA-2-Chat benefit from additional training with NEFTune., Comment: 25 pages, Code is available on Github: https://github.com/neelsjain/NEFTune

Prostate cancer (PCa) is a severe disease among men globally. It is important to identify PCa early and make a precise diagnosis for effective treatment. For PCa diagnosis, Multi-parametric magnetic resonance imaging (mpMRI) emerged as an invaluable imaging modality that offers a precise anatomical view of the prostate gland and its tissue structure. Deep learning (DL) models can enhance existing clinical systems and improve patient care by locating regions of interest for physicians. Recently, DL techniques have been employed to develop a pipeline for segmenting and classifying different cancer types. These studies show that DL can be used to increase diagnostic precision and give objective results without variability. This work uses well-known DL models for the classification and segmentation of mpMRI images to detect PCa. Our implementation involves four pipelines; Semantic DeepSegNet with ResNet50, DeepSegNet with recurrent neural network (RNN), U-Net with RNN, and U-Net with a long short-term memory (LSTM). Each segmentation model is paired with a different classifier to evaluate the performance using different metrics. The results of our experiments show that the pipeline that uses the combination of U-Net and the LSTM model outperforms all other combinations, excelling in both segmentation and classification tasks., Comment: Accepted at CISCON-2023

Motivated by the growing interest in correlation-robust stochastic optimization, we investigate stochastic selection problems beyond independence. Specifically, we consider the instructive case of pairwise-independent priors and matroid constraints. We obtain essentially-optimal bounds for contention resolution and prophet inequalities. The impetus for our work comes from the recent work of Caragiannis et al., who derived a constant-approximation for the single-choice prophet inequality with pairwise-independent priors. For general matroids, our results are tight and largely negative. For both contention resolution and prophet inequalities, our impossibility results hold for the full linear matroid over a finite field. We explicitly construct pairwise-independent distributions which rule out an omega(1/Rank)-balanced offline CRS and an omega(1/log Rank)-competitive prophet inequality against the (usual) oblivious adversary. For both results, we employ a generic approach for constructing pairwise-independent random vectors -- one which unifies and generalizes existing pairwise-independence constructions from the literature on universal hash functions and pseudorandomness. Specifically, our approach is based on our observation that random linear maps turn linear independence into stochastic independence. We then examine the class of matroids which satisfy the so-called partition property -- these include most common matroids encountered in optimization. We obtain positive results for both online contention resolution and prophet inequalities with pairwise-independent priors on such matroids, approximately matching the corresponding guarantees for fully independent priors. These algorithmic results hold against the almighty adversary for both problems., Comment: 44 pages, 2 figures, Accepted to STOC 24. In the metadata, mathematical expressions have been converted to plain text for simplicity and clarity

We present a single attention head in GPT-2 Small that has one main role across the entire training distribution. If components in earlier layers predict a certain token, and this token appears earlier in the context, the head suppresses it: we call this copy suppression. Attention Head 10.7 (L10H7) suppresses naive copying behavior which improves overall model calibration. This explains why multiple prior works studying certain narrow tasks found negative heads that systematically favored the wrong answer. We uncover the mechanism that the Negative Heads use for copy suppression with weights-based evidence and are able to explain 76.9% of the impact of L10H7 in GPT-2 Small. To the best of our knowledge, this is the most comprehensive description of the complete role of a component in a language model to date. One major effect of copy suppression is its role in self-repair. Self-repair refers to how ablating crucial model components results in downstream neural network parts compensating for this ablation. Copy suppression leads to self-repair: if an initial overconfident copier is ablated, then there is nothing to suppress. We show that self-repair is implemented by several mechanisms, one of which is copy suppression, which explains 39% of the behavior in a narrow task. Interactive visualisations of the copy suppression phenomena may be seen at our web app https://copy-suppression.streamlit.app/

The placenta plays a crucial role in fetal development. Automated 3D placenta segmentation from fetal EPI MRI holds promise for advancing prenatal care. This paper proposes an effective semi-supervised learning method for improving placenta segmentation in fetal EPI MRI time series. We employ consistency regularization loss that promotes consistency under spatial transformation of the same image and temporal consistency across nearby images in a time series. The experimental results show that the method improves the overall segmentation accuracy and provides better performance for outliers and hard samples. The evaluation also indicates that our method improves the temporal coherency of the prediction, which could lead to more accurate computation of temporal placental biomarkers. This work contributes to the study of the placenta and prenatal clinical decision-making. Code is available at https://github.com/firstmover/cr-seg.

Infrared (IR) dimension function $d_\text{IR}(\lambda)$ characterizes the space effectively utilized by QCD quarks at Dirac scale $\lambda$, and indirectly the space occupied by glue fields. It was proposed that its non-analytic behavior in thermal infrared phase reflects the separation of QCD system into an IR component and an independent bulk. Here we study the ``plateau modes" in IR component, whose dimensional properties were puzzling. Indeeed, in the recent metal-to-critical scenario of transition to IR phase, this low-dimensional plateau connects the Anderson-like mobility edge $\lambda_\text{IR}=0$ in Dirac spectrum with mobility edges $\pm \lambda_\text{A}$. For this structure to be truly Anderson-like, plateau modes have to be exponentially localized, implying that both the effective distances $L_\text{eff} \propto L^\gamma$ and the effective volumes $V_\text{eff} \propto L^{d_\text{IR}}$ in these modes grow slower than any positive power of IR cutoff $L$. Although $\gamma=0$ was confirmed in the plateau, it was found that $d_\text{IR}\approx 1$. Here we apply the recently proposed multidimension technique to the problem. We conclude that a plateau mode of pure-glue QCD at UV cutoff $a \!=\! 0.085\,$fm occupies a subvolume of IR dimension zero with probability at least 0.9999, substantiating this aspect of metal-to-critical scenario to a respective degree., Comment: 5 pages, 7 figures; v2: minor improvements, published version

Software testing is a crucial aspect of software development, and the creation of high-quality tests that adhere to best practices is essential for effective maintenance. Recently, Large Language Models (LLMs) have gained popularity for code generation, including the automated creation of test cases. However, these LLMs are often trained on vast amounts of publicly available code, which may include test cases that do not adhere to best practices and may even contain test smells (anti-patterns). To address this issue, we propose a novel technique called Reinforcement Learning from Static Quality Metrics (RLSQM). To begin, we analyze the anti-patterns generated by the LLM and show that LLMs can generate undesirable test smells. Thus, we train specific reward models for each static quality metric, then utilize Proximal Policy Optimization (PPO) to train models for optimizing a single quality metric at a time. Furthermore, we amalgamate these rewards into a unified reward model aimed at capturing different best practices and quality aspects of tests. By comparing RL-trained models with those trained using supervised learning, we provide insights into how reliably utilize RL to improve test generation quality and into the effects of various training strategies. Our experimental results demonstrate that the RL-optimized model consistently generated high-quality test cases compared to the base LLM, improving the model by up to 21%, and successfully generates nearly 100% syntactically correct code. RLSQM also outperformed GPT-4 on four out of seven metrics. This represents a significant step towards enhancing the overall efficiency and reliability of software testing through Reinforcement Learning and static quality metrics. Our data are available at this link: https://figshare.com/s/ded476c8d4c221222849.

In recent years, the increasing threat of devastating wildfires has underscored the need for effective prescribed fire management. Process-based computer simulations have traditionally been employed to plan prescribed fires for wildfire prevention. However, even simplified process models like QUIC-Fire are too compute-intensive to be used for real-time decision-making, especially when weather conditions change rapidly. Traditional ML methods used for fire modeling offer computational speedup but struggle with physically inconsistent predictions, biased predictions due to class imbalance, biased estimates for fire spread metrics (e.g., burned area, rate of spread), and generalizability in out-of-distribution wind conditions. This paper introduces a novel machine learning (ML) framework that enables rapid emulation of prescribed fires while addressing these concerns. By incorporating domain knowledge, the proposed method helps reduce physical inconsistencies in fuel density estimates in data-scarce scenarios. To overcome the majority class bias in predictions, we leverage pre-existing source domain data to augment training data and learn the spread of fire more effectively. Finally, we overcome the problem of biased estimation of fire spread metrics by incorporating a hierarchical modeling structure to capture the interdependence in fuel density and burned area. Notably, improvement in fire metric (e.g., burned area) estimates offered by our framework makes it useful for fire managers, who often rely on these fire metric estimates to make decisions about prescribed burn management. Furthermore, our framework exhibits better generalization capabilities than the other ML-based fire modeling methods across diverse wind conditions and ignition patterns.

In this work, we build on our method for manipulating unknown objects via contact configuration regulation: the estimation and control of the location, geometry, and mode of all contacts between the robot, object, and environment. We further develop our estimator and controller to enable manipulation through more complex contact interactions, including intermittent contact between the robot/object, and multiple contacts between the object/environment. In addition, we support a larger set of contact geometries at each interface. This is accomplished through a factor graph based estimation framework that reasons about the complementary kinematic and wrench constraints of contact to predict the current contact configuration. We are aided by the incorporation of a limited amount of visual feedback; which when combined with the available F/T sensing and robot proprioception, allows us to differentiate contact modes that were previously indistinguishable. We implement this revamped framework on our manipulation platform, and demonstrate that it allows the robot to perform a wider set of manipulation tasks. This includes, using a wall as a support to re-orient an object, or regulating the contact geometry between the object and the ground. Finally, we conduct ablation studies to understand the contributions from visual and tactile feedback in our manipulation framework. Our code can be found at: https://github.com/mcubelab/pbal.

Building an interactive AI assistant that can perceive, reason, and collaborate with humans in the real world has been a long-standing pursuit in the AI community. This work is part of a broader research effort to develop intelligent agents that can interactively guide humans through performing tasks in the physical world. As a first step in this direction, we introduce HoloAssist, a large-scale egocentric human interaction dataset, where two people collaboratively complete physical manipulation tasks. The task performer executes the task while wearing a mixed-reality headset that captures seven synchronized data streams. The task instructor watches the performer's egocentric video in real time and guides them verbally. By augmenting the data with action and conversational annotations and observing the rich behaviors of various participants, we present key insights into how human assistants correct mistakes, intervene in the task completion procedure, and ground their instructions to the environment. HoloAssist spans 166 hours of data captured by 350 unique instructor-performer pairs. Furthermore, we construct and present benchmarks on mistake detection, intervention type prediction, and hand forecasting, along with detailed analysis. We expect HoloAssist will provide an important resource for building AI assistants that can fluidly collaborate with humans in the real world. Data can be downloaded at https://holoassist.github.io/., Comment: ICCV 2023

Mechanistic interpretability seeks to understand the internal mechanisms of machine learning models, where localization -- identifying the important model components -- is a key step. Activation patching, also known as causal tracing or interchange intervention, is a standard technique for this task (Vig et al., 2020), but the literature contains many variants with little consensus on the choice of hyperparameters or methodology. In this work, we systematically examine the impact of methodological details in activation patching, including evaluation metrics and corruption methods. In several settings of localization and circuit discovery in language models, we find that varying these hyperparameters could lead to disparate interpretability results. Backed by empirical observations, we give conceptual arguments for why certain metrics or methods may be preferred. Finally, we provide recommendations for the best practices of activation patching going forwards., Comment: 27 pages. ICLR 2024

We propose a new paradigm to automatically generate training data with accurate labels at scale using the text-to-image synthesis frameworks (e.g., DALL-E, Stable Diffusion, etc.). The proposed approach1 decouples training data generation into foreground object generation, and contextually coherent background generation. To generate foreground objects, we employ a straightforward textual template, incorporating the object class name as input prompts. This is fed into a text-to-image synthesis framework, producing various foreground images set against isolated backgrounds. A foreground-background segmentation algorithm is then used to generate foreground object masks. To generate context images, we begin by creating language descriptions of the context. This is achieved by applying an image captioning method to a small set of images representing the desired context. These textual descriptions are then transformed into a diverse array of context images via a text-to-image synthesis framework. Subsequently, we composite these with the foreground object masks produced in the initial step, utilizing a cut-and-paste method, to formulate the training data. We demonstrate the advantages of our approach on five object detection and segmentation datasets, including Pascal VOC and COCO. We found that detectors trained solely on synthetic data produced by our method achieve performance comparable to those trained on real data (Fig. 1). Moreover, a combination of real and synthetic data yields even much better results. Further analysis indicates that the synthetic data distribution complements the real data distribution effectively. Additionally, we emphasize the compositional nature of our data generation approach in out-of-distribution and zero-shot data generation scenarios. We open-source our code at https://github.com/gyhandy/Text2Image-for-Detection, Comment: Code in https://github.com/gyhandy/Text2Image-for-Detection

Demographics, Social determinants of health, and family history documented in the unstructured text within the electronic health records are increasingly being studied to understand how this information can be utilized with the structured data to improve healthcare outcomes. After the GPT models were released, many studies have applied GPT models to extract this information from the narrative clinical notes. Different from the existing work, our research focuses on investigating the zero-shot learning on extracting this information together by providing minimum information to the GPT model. We utilize de-identified real-world clinical notes annotated for demographics, various social determinants, and family history information. Given that the GPT model might provide text different from the text in the original data, we explore two sets of evaluation metrics, including the traditional NER evaluation metrics and semantic similarity evaluation metrics, to completely understand the performance. Our results show that the GPT-3.5 method achieved an average of 0.975 F1 on demographics extraction, 0.615 F1 on social determinants extraction, and 0.722 F1 on family history extraction. We believe these results can be further improved through model fine-tuning or few-shots learning. Through the case studies, we also identified the limitations of the GPT models, which need to be addressed in future research., Comment: 5 pages, 4 figures

How do sequence models represent their decision-making process? Prior work suggests that Othello-playing neural network learned nonlinear models of the board state (Li et al., 2023). In this work, we provide evidence of a closely related linear representation of the board. In particular, we show that probing for "my colour" vs. "opponent's colour" may be a simple yet powerful way to interpret the model's internal state. This precise understanding of the internal representations allows us to control the model's behaviour with simple vector arithmetic. Linear representations enable significant interpretability progress, which we demonstrate with further exploration of how the world model is computed.

As Large Language Models quickly become ubiquitous, it becomes critical to understand their security vulnerabilities. Recent work shows that text optimizers can produce jailbreaking prompts that bypass moderation and alignment. Drawing from the rich body of work on adversarial machine learning, we approach these attacks with three questions: What threat models are practically useful in this domain? How do baseline defense techniques perform in this new domain? How does LLM security differ from computer vision? We evaluate several baseline defense strategies against leading adversarial attacks on LLMs, discussing the various settings in which each is feasible and effective. Particularly, we look at three types of defenses: detection (perplexity based), input preprocessing (paraphrase and retokenization), and adversarial training. We discuss white-box and gray-box settings and discuss the robustness-performance trade-off for each of the defenses considered. We find that the weakness of existing discrete optimizers for text, combined with the relatively high costs of optimization, makes standard adaptive attacks more challenging for LLMs. Future research will be needed to uncover whether more powerful optimizers can be developed, or whether the strength of filtering and preprocessing defenses is greater in the LLMs domain than it has been in computer vision., Comment: 12 pages

We explored the potential applications of various Quantum Algorithms for stock price prediction by conducting a series of experimental simulations using both Classical as well as Quantum Hardware. Firstly, we extracted various stock price indicators, such as Moving Averages (MA), Average True Range (ATR), and Aroon, to gain insights into market trends and stock price movements. Next, we employed Quantum Annealing (QA) for feature selection and Principal Component Analysis (PCA) for dimensionality reduction. Further, we transformed the stock price prediction task essentially into a classification problem. We trained the Quantum Support Vector Machine (QSVM) to predict price movements (whether up or down) contrasted their performance with classical models and analyzed their accuracy on a dataset formulated using Quantum Annealing and PCA individually. We focused on the stock price prediction and binary classification of stock prices for four different companies, namely Apple, Visa, Johnson and Jonson, and Honeywell. We primarily used the real-time stock data of the raw stock prices of these companies. We compared various Quantum Computing techniques with their classical counterparts in terms of accuracy and F-score of the prediction model. Through these experimental simulations, we shed light on the potential advantages and limitations of Quantum Algorithms in stock price prediction and contribute to the growing body of knowledge at the intersection of Quantum Computing and Finance., Comment: RASSE2023 IEEE

The advent of large language models (LLMs) and their adoption by the legal community has given rise to the question: what types of legal reasoning can LLMs perform? To enable greater study of this question, we present LegalBench: a collaboratively constructed legal reasoning benchmark consisting of 162 tasks covering six different types of legal reasoning. LegalBench was built through an interdisciplinary process, in which we collected tasks designed and hand-crafted by legal professionals. Because these subject matter experts took a leading role in construction, tasks either measure legal reasoning capabilities that are practically useful, or measure reasoning skills that lawyers find interesting. To enable cross-disciplinary conversations about LLMs in the law, we additionally show how popular legal frameworks for describing legal reasoning -- which distinguish between its many forms -- correspond to LegalBench tasks, thus giving lawyers and LLM developers a common vocabulary. This paper describes LegalBench, presents an empirical evaluation of 20 open-source and commercial LLMs, and illustrates the types of research explorations LegalBench enables., Comment: 143 pages, 79 tables, 4 figures

We study the combinatorial contract design problem, introduced and studied by Dutting et. al. (2021, 2022), in both the single and multi-agent settings. Prior work has examined the problem when the principal's utility function is submodular in the actions chosen by the agent(s). We complement this emerging literature with an examination of the problem when the principal's utility is supermodular. In the single-agent setting, we obtain a strongly polynomial time algorithm for the optimal contract. This stands in contrast to the NP-hardness of the problem with submodular principal utility due to Dutting et. al. (2021). This result has two technical components, the first of which applies beyond supermodular or submodular utilities. This result strengthens and simplifies analogous enumeration algorithms from Dutting et. al. (2021), and applies to any nondecreasing valuation function for the principal. Second, we show that supermodular valuations lead to a polynomial number of breakpoints, analogous to a similar result by Dutting et. al. (2021) for gross substitutes valuations. In the multi-agent setting, we obtain a mixed bag of positive and negative results. First, we show that it is NP-hard to obtain any finite multiplicative approximation, or an additive FPTAS. This stands in contrast to the submodular case, where efficient computation of approximately optimal contracts was shown by Dutting et. al. (2022). Second, we derive an additive PTAS for the problem in the instructive special case of graph-based supermodular valuations, and equal costs. En-route to this result, we discover an intimate connection between the multi-agent contract problem and the notorious k-densest subgraph problem. We build on and combine techniques from the literature on dense subgraph problems to obtain our additive PTAS., Comment: 31 pages, 2 figures

For a large class of random constraint satisfaction problems (CSP), deep but non-rigorous theory from statistical physics predict the location of the sharp satisfiability transition. The works of Ding, Sly, Sun (2014, 2016) and Coja-Oghlan, Panagiotou (2014) established the satisfiability threshold for random regular $k$-NAE-SAT, random $k$-SAT, and random regular $k$-SAT for large enough $k\geq k_0$ where $k_0$ is a large non-explicit constant. Establishing the same for small values of $k\geq 3$ remains an important open problem in the study of random CSPs. In this work, we study two closely related models of random CSPs, namely the $2$-coloring on random $d$-regular $k$-uniform hypergraphs and the random $d$-regular $k$-NAE-SAT model. For every $k\geq 3$, we prove that there is an explicit $d_{\ast}(k)$ which gives a satisfiability upper bound for both of the models. Our upper bound $d_{\ast}(k)$ for $k\geq 3$ matches the prediction from statistical physics for the hypergraph $2$-coloring by Dall'Asta, Ramezanpour, Zecchina (2008), thus conjectured to be sharp. Moreover, $d_{\ast}(k)$ coincides with the satisfiability threshold of random regular $k$-NAE-SAT for large enough $k\geq k_0$ by Ding, Sly, Sun (2014)., Comment: 23 pages, 1 table

Code coverage is a widely used metric for quantifying the extent to which program elements, such as statements or branches, are executed during testing. Calculating code coverage is resource-intensive, requiring code building and execution with additional overhead for the instrumentation. Furthermore, computing coverage of any snippet of code requires the whole program context. Using Machine Learning to amortize this expensive process could lower the cost of code coverage by requiring only the source code context, and the task of code coverage prediction can be a novel benchmark for judging the ability of models to understand code. We propose a novel benchmark task called Code Coverage Prediction for Large Language Models (LLMs). We formalize this task to evaluate the capability of LLMs in understanding code execution by determining which lines of a method are executed by a given test case and inputs. We curate and release a dataset we call COVERAGEEVAL by executing tests and code from the HumanEval dataset and collecting code coverage information. We report the performance of four state-of-the-art LLMs used for code-related tasks, including OpenAI's GPT-4 and GPT-3.5-Turbo, Google's BARD, and Anthropic's Claude, on the Code Coverage Prediction task. Finally, we argue that code coverage as a metric and pre-training data source are valuable for overall LLM performance on software engineering tasks.

Language Models today provide a high accuracy across a large number of downstream tasks. However, they remain susceptible to adversarial attacks, particularly against those where the adversarial examples maintain considerable similarity to the original text. Given the multilingual nature of text, the effectiveness of adversarial examples across translations and how machine translations can improve the robustness of adversarial examples remain largely unexplored. In this paper, we present a comprehensive study on the robustness of current text adversarial attacks to round-trip translation. We demonstrate that 6 state-of-the-art text-based adversarial attacks do not maintain their efficacy after round-trip translation. Furthermore, we introduce an intervention-based solution to this problem, by integrating Machine Translation into the process of adversarial example generation and demonstrating increased robustness to round-trip translation. Our results indicate that finding adversarial examples robust to translation can help identify the insufficiency of language models that is common across languages, and motivate further research into multilingual adversarial attacks., Comment: Published at International Conference on Acoustics, Speech, and Signal Processing (ICASSP) 2023

Recent work has shown that language models' (LMs) prompt-based learning capabilities make them well suited for automating data labeling in domains where manual annotation is expensive. The challenge is that while writing an initial prompt is cheap, improving a prompt is costly -- practitioners often require significant labeled data in order to evaluate the impact of prompt modifications. Our work asks whether it is possible to improve prompt-based learning without additional labeled data. We approach this problem by attempting to modify the predictions of a prompt, rather than the prompt itself. Our intuition is that accurate predictions should also be consistent: samples which are similar under some feature representation should receive the same prompt prediction. We propose Embroid, a method which computes multiple representations of a dataset under different embedding functions, and uses the consistency between the LM predictions for neighboring samples to identify mispredictions. Embroid then uses these neighborhoods to create additional predictions for each sample, and combines these predictions with a simple latent variable graphical model in order to generate a final corrected prediction. In addition to providing a theoretical analysis of Embroid, we conduct a rigorous empirical evaluation across six different LMs and up to 95 different tasks. We find that (1) Embroid substantially improves performance over original prompts (e.g., by an average of 7.3 points on GPT-JT), (2) also realizes improvements for more sophisticated prompting strategies (e.g., chain-of-thought), and (3) can be specialized to domains like law through the embedding functions., Comment: 38 pages, 22 figures, 8 tables

\emph{Circuit analysis} is a promising technique for understanding the internal mechanisms of language models. However, existing analyses are done in small models far from the state of the art. To address this, we present a case study of circuit analysis in the 70B Chinchilla model, aiming to test the scalability of circuit analysis. In particular, we study multiple-choice question answering, and investigate Chinchilla's capability to identify the correct answer \emph{label} given knowledge of the correct answer \emph{text}. We find that the existing techniques of logit attribution, attention pattern visualization, and activation patching naturally scale to Chinchilla, allowing us to identify and categorize a small set of `output nodes' (attention heads and MLPs). We further study the `correct letter' category of attention heads aiming to understand the semantics of their features, with mixed results. For normal multiple-choice question answers, we significantly compress the query, key and value subspaces of the head without loss of performance when operating on the answer labels for multiple-choice questions, and we show that the query and key subspaces represent an `Nth item in an enumeration' feature to at least some extent. However, when we attempt to use this explanation to understand the heads' behaviour on a more general distribution including randomized answer labels, we find that it is only a partial explanation, suggesting there is more to learn about the operation of `correct letter' heads on multiple choice question answering.

The advancement of biomedical research heavily relies on access to large amounts of medical data. In the case of histopathology, Whole Slide Images (WSI) and clinicopathological information are valuable for developing Artificial Intelligence (AI) algorithms for Digital Pathology (DP). Transferring medical data "as open as possible" enhances the usability of the data for secondary purposes but poses a risk to patient privacy. At the same time, existing regulations push towards keeping medical data "as closed as necessary" to avoid re-identification risks. Generally, these legal regulations require the removal of sensitive data but do not consider the possibility of data linkage attacks due to modern image-matching algorithms. In addition, the lack of standardization in DP makes it harder to establish a single solution for all formats of WSIs. These challenges raise problems for bio-informatics researchers in balancing privacy and progress while developing AI algorithms. This paper explores the legal regulations and terminologies for medical data-sharing. We review existing approaches and highlight challenges from the histopathological perspective. We also present a data-sharing guideline for histological data to foster multidisciplinary research and education., Comment: Accepted to FAIEMA 2023

Neural network prediction probabilities and accuracy are often only weakly-correlated. Inherent label ambiguity in training data for image segmentation aggravates such miscalibration. We show that logit consistency across stochastic transformations acts as a spatially varying regularizer that prevents overconfident predictions at pixels with ambiguous labels. Our boundary-weighted extension of this regularizer provides state-of-the-art calibration for prostate and heart MRI segmentation., Comment: Accepted for publication at MICCAI 2023

Star-convex shapes arise across bio-microscopy and radiology in the form of nuclei, nodules, metastases, and other units. Existing instance segmentation networks for such structures train on densely labeled instances for each dataset, which requires substantial and often impractical manual annotation effort. Further, significant reengineering or finetuning is needed when presented with new datasets and imaging modalities due to changes in contrast, shape, orientation, resolution, and density. We present AnyStar, a domain-randomized generative model that simulates synthetic training data of blob-like objects with randomized appearance, environments, and imaging physics to train general-purpose star-convex instance segmentation networks. As a result, networks trained using our generative model do not require annotated images from unseen datasets. A single network trained on our synthesized data accurately 3D segments C. elegans and P. dumerilii nuclei in fluorescence microscopy, mouse cortical nuclei in micro-CT, zebrafish brain nuclei in EM, and placental cotyledons in human fetal MRI, all without any retraining, finetuning, transfer learning, or domain adaptation. Code is available at https://github.com/neel-dey/AnyStar., Comment: Code available at https://github.com/neel-dey/AnyStar

Digitized histopathology glass slides, known as Whole Slide Images (WSIs), are often several gigapixels large and contain sensitive metadata information, which makes distributed processing unfeasible. Moreover, artifacts in WSIs may result in unreliable predictions when directly applied by Deep Learning (DL) algorithms. Therefore, preprocessing WSIs is beneficial, e.g., eliminating privacy-sensitive information, splitting a gigapixel medical image into tiles, and removing the diagnostically irrelevant areas. This work proposes a cloud service to parallelize the preprocessing pipeline for large medical images. The data and model parallelization will not only boost the end-to-end processing efficiency for histological tasks but also secure the reconstruction of WSI by randomly distributing tiles across processing nodes. Furthermore, the initial steps of the pipeline will be integrated into the Jupyter-based Virtual Research Environment (VRE) to enable image owners to configure and automate the execution process based on resource allocation., Comment: Accepted at IEEE ICDH 2023

Extracorporeal membrane oxygenation (ECMO) is an essential life-supporting modality for COVID-19 patients who are refractory to conventional therapies. However, the proper treatment decision has been the subject of significant debate and it remains controversial about who benefits from this scarcely available and technically complex treatment option. To support clinical decisions, it is a critical need to predict the treatment need and the potential treatment and no-treatment responses. Targeting this clinical challenge, we propose Treatment Variational AutoEncoder (TVAE), a novel approach for individualized treatment analysis. TVAE is specifically designed to address the modeling challenges like ECMO with strong treatment selection bias and scarce treatment cases. TVAE conceptualizes the treatment decision as a multi-scale problem. We model a patient's potential treatment assignment and the factual and counterfactual outcomes as part of their intrinsic characteristics that can be represented by a deep latent variable model. The factual and counterfactual prediction errors are alleviated via a reconstruction regularization scheme together with semi-supervision, and the selection bias and the scarcity of treatment cases are mitigated by the disentangled and distribution-matched latent space and the label-balancing generative strategy. We evaluate TVAE on two real-world COVID-19 datasets: an international dataset collected from 1651 hospitals across 63 countries, and a institutional dataset collected from 15 hospitals. The results show that TVAE outperforms state-of-the-art treatment effect models in predicting both the propensity scores and factual outcomes on heterogeneous COVID-19 datasets. Additional experiments also show TVAE outperforms the best existing models in individual treatment effect estimation on the synthesized IHDP benchmark dataset.

Performance bugs are non-functional bugs that can even manifest in well-tested commercial products. Fixing these performance bugs is an important yet challenging problem. In this work, we address this challenge and present a new approach called Retrieval-Augmented Prompt Generation (RAPGen). Given a code snippet with a performance issue, RAPGen first retrieves a prompt instruction from a pre-constructed knowledge-base of previous performance bug fixes and then generates a prompt using the retrieved instruction. It then uses this prompt on a Large Language Model (such as Codex) in zero-shot to generate a fix. We compare our approach with the various prompt variations and state of the art methods in the task of performance bug fixing. Our evaluation shows that RAPGen can generate performance improvement suggestions equivalent or better than a developer in ~60% of the cases, getting ~42% of them verbatim, in an expert-verified dataset of past performance changes made by C# developers.

With the rise of Large Language Models (LLMs) and their ubiquitous deployment in diverse domains, measuring language model behavior on realistic data is imperative. For example, a company deploying a client-facing chatbot must ensure that the model will not respond to client requests with profanity. Current evaluations approach this problem using small, domain-specific datasets with human-curated labels. These evaluation sets are often sampled from a narrow and simplified distribution, and data sources can unknowingly be leaked into the training set which can lead to misleading evaluations. To bypass these drawbacks, we propose a framework for self-supervised evaluation of LLMs by analyzing their sensitivity or invariance to transformations on the input text. Self-supervised evaluation can directly monitor LLM behavior on datasets collected in the wild or streamed during live model deployment. We demonstrate self-supervised evaluation strategies for measuring closed-book knowledge, toxicity, and long-range context dependence, in addition to sensitivity to grammatical structure and tokenization errors. When comparisons to similar human-labeled benchmarks are available, we find strong correlations between self-supervised and human-supervised evaluations. The self-supervised paradigm complements current evaluation strategies that rely on labeled data., Comment: Code is available at https://github.com/neelsjain/BYOD. First two authors contributed equally. 21 pages, 22 figures

The recent development of science education leads educators to explore new teaching and learning methodologies and restructure classes and assignments to bring students' knowledge to the highest level of education by allowing learners to gain various skills that are reflected in their future. We are discussing how to develop and deliver effective online courses by personalizing the class and individualizing the assignments for learners to succeed in Physics. Various teaching and learning activities are aimed at effective learning and interaction to enhance student learning in this class structure. The students' performances and progress are analyzed in four large introductory Physics classes. Student registration and retention are well preserved by providing learners freedom of education, a sense of belonging, mindset-growing encouragements, effective feedback, problem-solving strategies, and one-to-one communications. Academic integrity is well-balanced by individualizing the assignments, personalizing class materials, breaking down problems, and investigating ChatGPT usage. The students' performances are analyzed by evaluating class performance in total grade distribution, missing assignments, and choice-based learning. Student progress is investigated by analyzing their commitment, respect, willingness to improve, and time management toward education. Overall, the class model demonstrates great performance, progress, and achievements that indicate a positive learning environment for all students throughout the semester. Hence, the effective model with personalized assignments, and other strategies that we share will be highly beneficial for science education., Comment: 12 pages, 10 figures

Advances in Large Language Models (LLMs) have led to remarkable capabilities, yet their inner mechanisms remain largely unknown. To understand these models, we need to unravel the functions of individual neurons and their contribution to the network. This paper introduces a novel automated approach designed to scale interpretability techniques across a vast array of neurons within LLMs, to make them more interpretable and ultimately safe. Conventional methods require examination of examples with strong neuron activation and manual identification of patterns to decipher the concepts a neuron responds to. We propose Neuron to Graph (N2G), an innovative tool that automatically extracts a neuron's behaviour from the dataset it was trained on and translates it into an interpretable graph. N2G uses truncation and saliency methods to emphasise only the most pertinent tokens to a neuron while enriching dataset examples with diverse samples to better encompass the full spectrum of neuron behaviour. These graphs can be visualised to aid researchers' manual interpretation, and can generate token activations on text for automatic validation by comparison with the neuron's ground truth activations, which we use to show that the model is better at predicting neuron activation than two baseline methods. We also demonstrate how the generated graph representations can be flexibly used to facilitate further automation of interpretability research, by searching for neurons with particular properties, or programmatically comparing neurons to each other to identify similar neurons. Our method easily scales to build graph representations for all neurons in a 6-layer Transformer model using a single Tesla T4 GPU, allowing for wide usability. We release the code and instructions for use at https://github.com/alexjfoote/Neuron2Graph.

Close hyperbolic encounters of black holes (BHs) generate certain Burst With Memory (BWM) events in the frequency windows of the operational, planned, and proposed gravitational wave (GW) observatories. We present detailed explorations of the detectable parameter space of such events that are relevant for the LIGO-Virgo-KAGRA and the International Pulsar Timing Array (IPTA) consortia. The underlying temporally evolving GW polarization states are adapted from Cho et al. [Phys. Rev. D 98, 024039 (2018)] and therefore incorporate general relativistic effects up to the third post-Newtonian order. Further, we provide a prescription to ensure the validity of our waveform family while describing close encounters. Preliminary investigations reveal that optimally placed BWM events should be visible to megaparsec distances for the existing ground-based observatories. In contrast, maturing IPTA datasets should be able to provide constraints on the occurrences of such hyperbolic encounters of supermassive BHs to gigaparsec distances., Comment: 19 pages, 11 figures, accepted for publication in Phys. Rev. D

Current text-to-image generation models often struggle to follow textual instructions, especially the ones requiring spatial reasoning. On the other hand, Large Language Models (LLMs), such as GPT-4, have shown remarkable precision in generating code snippets for sketching out text inputs graphically, e.g., via TikZ. In this work, we introduce Control-GPT to guide the diffusion-based text-to-image pipelines with programmatic sketches generated by GPT-4, enhancing their abilities for instruction following. Control-GPT works by querying GPT-4 to write TikZ code, and the generated sketches are used as references alongside the text instructions for diffusion models (e.g., ControlNet) to generate photo-realistic images. One major challenge to training our pipeline is the lack of a dataset containing aligned text, images, and sketches. We address the issue by converting instance masks in existing datasets into polygons to mimic the sketches used at test time. As a result, Control-GPT greatly boosts the controllability of image generation. It establishes a new state-of-art on the spatial arrangement and object positioning generation and enhances users' control of object positions, sizes, etc., nearly doubling the accuracy of prior models. Our work, as a first attempt, shows the potential for employing LLMs to enhance the performance in computer vision tasks.

Computational Pathology (CPATH) systems have the potential to automate diagnostic tasks. However, the artifacts on the digitized histological glass slides, known as Whole Slide Images (WSIs), may hamper the overall performance of CPATH systems. Deep Learning (DL) models such as Vision Transformers (ViTs) may detect and exclude artifacts before running the diagnostic algorithm. A simple way to develop robust and generalized ViTs is to train them on massive datasets. Unfortunately, acquiring large medical datasets is expensive and inconvenient, prompting the need for a generalized artifact detection method for WSIs. In this paper, we present a student-teacher recipe to improve the classification performance of ViT for the air bubbles detection task. ViT, trained under the student-teacher framework, boosts its performance by distilling existing knowledge from the high-capacity teacher model. Our best-performing ViT yields 0.961 and 0.911 F1-score and MCC, respectively, observing a 7% gain in MCC against stand-alone training. The proposed method presents a new perspective of leveraging knowledge distillation over transfer learning to encourage the use of customized transformers for efficient preprocessing pipelines in the CPATH systems., Comment: Accepted at PAKDD 2023

Software vulnerabilities bear enterprises significant costs. Despite extensive efforts in research and development of software vulnerability detection methods, uncaught vulnerabilities continue to put software owners and users at risk. Many current vulnerability detection methods require that code snippets can compile and build before attempting detection. This, unfortunately, introduces a long latency between the time a vulnerability is injected to the time it is removed, which can substantially increases the cost of fixing a vulnerability. We recognize that the current advances in machine learning can be used to detect vulnerable code patterns on syntactically incomplete code snippets as the developer is writing the code at EditTime. In this paper we present a practical system that leverages deep learning on a large-scale data set of vulnerable code patterns to learn complex manifestations of more than 250 vulnerability types and detect vulnerable code patterns at EditTime. We discuss zero-shot, few-shot, and fine-tuning approaches on state of the art pre-trained Large Language Models (LLMs). We show that in comparison with state of the art vulnerability detection models our approach improves the state of the art by 10%. We also evaluate our approach to detect vulnerability in auto-generated code by code LLMs. Evaluation on a benchmark of high-risk code scenarios shows a reduction of up to 90% vulnerability reduction.

Code execution is a fundamental aspect of programming language semantics that reflects the exact behavior of the code. However, most pre-trained models for code intelligence ignore the execution trace and only rely on source code and syntactic structures. In this paper, we investigate how well pre-trained models can understand and perform code execution. We develop a mutation-based data augmentation technique to create a large-scale and realistic Python dataset and task for code execution, which challenges existing models such as Codex. We then present CodeExecutor, a Transformer model that leverages code execution pre-training and curriculum learning to enhance its semantic comprehension. We evaluate CodeExecutor on code execution and show its promising performance and limitations. We also demonstrate its potential benefits for code intelligence tasks such as zero-shot code-to-code search and text-to-code generation. Our analysis provides insights into the learning and generalization abilities of pre-trained models for code execution., Comment: Accepted to the Findings of ACL 2023

Despite rapid adoption and deployment of large language models (LLMs), the internal computations of these models remain opaque and poorly understood. In this work, we seek to understand how high-level human-interpretable features are represented within the internal neuron activations of LLMs. We train $k$-sparse linear classifiers (probes) on these internal activations to predict the presence of features in the input; by varying the value of $k$ we study the sparsity of learned representations and how this varies with model scale. With $k=1$, we localize individual neurons which are highly relevant for a particular feature, and perform a number of case studies to illustrate general properties of LLMs. In particular, we show that early layers make use of sparse combinations of neurons to represent many features in superposition, that middle layers have seemingly dedicated neurons to represent higher-level contextual features, and that increasing scale causes representational sparsity to increase on average, but there are multiple types of scaling dynamics. In all, we probe for over 100 unique features comprising 10 different categories in 7 different models spanning 70 million to 6.9 billion parameters.

The wideband timing technique enables the high-precision simultaneous estimation of pulsar Times of Arrival (ToAs) and Dispersion Measures (DMs) while effectively modeling frequency-dependent profile evolution. We present two novel independent methods that extend the standard wideband technique to handle simultaneous multi-band pulsar data incorporating profile evolution over a larger frequency span to estimate DMs and ToAs with enhanced precision. We implement the wideband likelihood using the libstempo python interface to perform wideband timing in the tempo2 framework. We present the application of these techniques to the dataset of fourteen millisecond pulsars observed simultaneously in Band 3 (300 - 500 MHz) and Band 5 (1260 - 1460 MHz) of the upgraded Giant Metrewave Radio Telescope (uGMRT) with a large band gap of 760 MHz as a part of the Indian Pulsar Timing Array (InPTA) campaign. We achieve increased ToA and DM precision and sub-microsecond root mean square post-fit timing residuals by combining simultaneous multi-band pulsar observations done in non-contiguous bands for the first time using our novel techniques., Comment: Published in MNRAS

Understanding the function of individual neurons within language models is essential for mechanistic interpretability research. We propose $\textbf{Neuron to Graph (N2G)}$, a tool which takes a neuron and its dataset examples, and automatically distills the neuron's behaviour on those examples to an interpretable graph. This presents a less labour intensive approach to interpreting neurons than current manual methods, that will better scale these methods to Large Language Models (LLMs). We use truncation and saliency methods to only present the important tokens, and augment the dataset examples with more diverse samples to better capture the extent of neuron behaviour. These graphs can be visualised to aid manual interpretation by researchers, but can also output token activations on text to compare to the neuron's ground truth activations for automatic validation. N2G represents a step towards scalable interpretability methods by allowing us to convert neurons in an LLM to interpretable representations of measurable quality., Comment: To be published at ICLR 2023 Workshop on Trustworthy and Reliable Large-Scale Machine Learning Models

In our work, we propose a novel yet simple approach to obtain an adaptive learning rate for gradient-based descent methods on classification tasks. Instead of the traditional approach of selecting adaptive learning rates via the decayed expectation of gradient-based terms, we use the angle between the current gradient and the new gradient: this new gradient is computed from the direction orthogonal to the current gradient, which further helps us in determining a better adaptive learning rate based on angle history, thereby, leading to relatively better accuracy compared to the existing state-of-the-art optimizers. On a wide variety of benchmark datasets with prominent image classification architectures such as ResNet, DenseNet, EfficientNet, and VGG, we find that our method leads to the highest accuracy in most of the datasets. Moreover, we prove that our method is convergent., Comment: 8 pages, 7 figures