arXiv subjects

Lav R. Varshney

Publications and source records attributed to Lav R. Varshney.

At least 73 records · Page 4Linked to original sources

arXiv2024

Learning from One and Only One Shot

Humans can generalize from only a few examples and from little pretraining on similar tasks. Yet, machine learning (ML) typically requires large data to learn or pre-learn to transfer. Motivated by nativism and artificial general intelligence, we directly model human-innate priors in abstract visual tasks such as character and doodle recognition. This yields a white-box model that learns general-appearance similarity by mimicking how humans naturally ``distort'' an object at first sight. Using just nearest-neighbor classification on this cognitively-inspired similarity space, we achieve human-level recognition with only $1$--$10$ examples per class and no pretraining. This differs from few-shot learning that uses massive pretraining. In the tiny-data regime of MNIST, EMNIST, Omniglot, and QuickDraw benchmarks, we outperform both modern neural networks and classical ML. For unsupervised learning, by learning the non-Euclidean, general-appearance similarity space in a $k$-means style, we achieve multifarious visual realizations of abstract concepts by generating human-intuitive archetypes as cluster centroids.

cs.CV↗

arXiv2024

Nonstationary Reinforcement Learning with Linear Function Approximation

We consider reinforcement learning (RL) in episodic Markov decision processes (MDPs) with linear function approximation under drifting environment. Specifically, both the reward and state transition functions can evolve over time but their total variations do not exceed a $\textit{variation budget}$. We first develop $\texttt{LSVI-UCB-Restart}$ algorithm, an optimistic modification of least-squares value iteration with periodic restart, and bound its dynamic regret when variation budgets are known. Then we propose a parameter-free algorithm $\texttt{Ada-LSVI-UCB-Restart}$ that extends to unknown variation budgets. We also derive the first minimax dynamic regret lower bound for nonstationary linear MDPs and as a byproduct establish a minimax regret lower bound for linear MDPs unsolved by Jin et al. (2020). Finally, we provide numerical experiments to demonstrate the effectiveness of our proposed algorithms.

cs.LG↗

arXiv2024

Semantic Compression with Information Lattice Learning

Data-driven artificial intelligence (AI) techniques are becoming prominent for learning in support of data compression, but are focused on standard problems such as text compression. To instead address the emerging problem of semantic compression, we argue that the lattice theory of information is particularly expressive and mathematically precise in capturing notions of abstraction as a form of lossy semantic compression. As such, we demonstrate that a novel AI technique called information lattice learning, originally developed for knowledge discovery and creativity, is powerful for learning to compress in a semantically-meaningful way. The lattice structure further implies the optimality of group codes and the successive refinement property for progressive transmission.

cs.IT↗

arXiv2024

A Meta-Learning Perspective on Transformers for Causal Language Modeling

The Transformer architecture has become prominent in developing large causal language models. However, mechanisms to explain its capabilities are not well understood. Focused on the training process, here we establish a meta-learning view of the Transformer architecture when trained for the causal language modeling task, by explicating an inner optimization process within the Transformer. Further, within the inner optimization, we discover and theoretically analyze a special characteristic of the norms of learned token representations within Transformer-based causal language models. Our analysis is supported by experiments in various settings.

cs.LG↗

arXiv2024

Transformer-based Causal Language Models Perform Clustering

Even though large language models (LLMs) have demonstrated remarkable capability in solving various natural language tasks, the capability of an LLM to follow human instructions is still a concern. Recent works have shown great improvements in the instruction-following capability via additional training for instruction-following tasks. However, the mechanisms responsible for effective instruction-following capabilities remain inadequately understood. Here, we introduce a simplified instruction-following task and use synthetic datasets to analyze a Transformer-based causal language model. Our findings suggest that the model learns task-specific information by clustering data within its hidden space, with this clustering process evolving dynamically during learning. We also demonstrate how this phenomenon assists the model in handling unseen instances, and validate our results in a more realistic setting. Furthermore, we present inspired applications regarding pre-training and alignment.

cs.CL↗

arXiv2024

Federated Learning via Lattice Joint Source-Channel Coding

This paper introduces a universal federated learning framework that enables over-the-air computation via digital communications, using a new joint source-channel coding scheme. Without relying on channel state information at devices, this scheme employs lattice codes to both quantize model parameters and exploit interference from the devices. A novel two-layer receiver structure at the server is designed to reliably decode an integer combination of the quantized model parameters as a lattice point for the purpose of aggregation. Numerical experiments validate the effectiveness of the proposed scheme. Even with the challenges posed by channel conditions and device heterogeneity, the proposed scheme markedly surpasses other over-the-air FL strategies.

cs.IT↗

arXiv2023

Designing Discontinuities

Discontinuities can be fairly arbitrary but also cause a significant impact on outcomes in larger systems. Indeed, their arbitrariness is why they have been used to infer causal relationships among variables in numerous settings. Regression discontinuity from econometrics assumes the existence of a discontinuous variable that splits the population into distinct partitions to estimate the causal effects of a given phenomenon. Here we consider the design of partitions for a given discontinuous variable to optimize a certain effect previously studied using regression discontinuity. To do so, we propose a quantization-theoretic approach to optimize the effect of interest, first learning the causal effect size of a given discontinuous variable and then applying dynamic programming for optimal quantization design of discontinuities to balance the gain and loss in that effect size. We also develop a computationally-efficient reinforcement learning algorithm for the dynamic programming formulation of optimal quantization. We demonstrate our approach by designing optimal time zone borders for counterfactuals of social capital, social mobility, and health. This is based on regression discontinuity analyses we perform on novel data, which may be of independent empirical interest.

cs.IT↗

arXiv2023

Optimal Recovery for Causal Inference

Problems in causal inference can be fruitfully addressed using signal processing techniques. As an example, it is crucial to successfully quantify the causal effects of an intervention to determine whether the intervention achieved desired outcomes. We present a new geometric signal processing approach to classical synthetic control called ellipsoidal optimal recovery (EOpR), for estimating the unobservable outcome of a treatment unit. EOpR provides policy evaluators with both worst-case and typical outcomes to help in decision making. It is an approximation-theoretic technique that relates to the theory of principal components, which recovers unknown observations given a learned signal class and a set of known observations. We show EOpR can improve pre-treatment fit and mitigate bias of the post-treatment estimate relative to other methods in causal inference. Beyond recovery of the unit of interest, an advantage of EOpR is that it produces worst-case limits over the estimates produced. We assess our approach on artificially-generated data, on datasets commonly used in the econometrics literature, and in the context of the COVID-19 pandemic, showing better performance than baseline techniques

stat.ME↗

arXiv2023

Muslim-Violence Bias Persists in Debiased GPT Models

Abid et al. (2021) showed a tendency in GPT-3 to generate mostly violent completions when prompted about Muslims, compared with other religions. Two pre-registered replication attempts found few violent completions and only a weak anti-Muslim bias in the more recent InstructGPT, fine-tuned to eliminate biased and toxic outputs. However, more pre-registered experiments showed that using common names associated with the religions in prompts increases several-fold the rate of violent completions, revealing a significant second-order anti-Muslim bias. ChatGPT showed a bias many times stronger regardless of prompt format, suggesting that the effects of debiasing were reduced with continued model development. Our content analysis revealed religion-specific themes containing offensive stereotypes across all experiments. Our results show the need for continual de-biasing of models in ways that address both explicit and higher-order associations.

cs.CL↗

arXiv2023

Dynamic Resource Allocation to Minimize Concave Costs of Shortfalls

We study a resource allocation problem over time, where a finite (random) resource needs to be distributed among a set of users at each time instant. Shortfalls in the resource allocated result in user dissatisfaction, which we model as an increasing function of the long-term average shortfall for each user. In many scenarios such as wireless multimedia streaming, renewable energy grid, or supply chain logistics, a natural choice for this cost function turns out to be concave, rather than usual convex cost functions. We consider minimizing the (normalized) cumulative cost across users. Depending on whether users' mean consumption rates are known or unknown, this problem can be reduced to two different structured non-convex problems. The "known" case is a concave minimization problem subject to a linear constraint. By exploiting a well-chosen linearization of the cost functions, we solve this provably within $\mathcal{O}\left(\frac{1}{m}\right)$ of the optimum, in $\mathcal{O}\left(m \log{m}\right)$ time, where $m$ is the number of users in the system. In the "unknown" case, we are faced with minimizing the sum of functions that are concave on part of the domain and convex on the rest, subject to a linear constraint. We present a provably exact algorithm when the cost functions and prior distributions on mean consumption are the same across all users.

math.OC↗

arXiv2023

Limits of Fault-Tolerance on Resource-Constrained Quantum Circuits for Classical Problems

Existing lower bounds on redundancy in fault-tolerant quantum circuits are applicable when both the input and the intended output are quantum states. These bounds may not necessarily hold, however, when the input and the intended output are classical bits, as in the Deutsch-Jozsa, Grover, or Shor algorithms. Here we show that indeed, noise thresholds obtained from existing bounds do not apply to a simple fault-tolerant implementation of the Deutsch-Jozsa algorithm. Then we obtain the first lower bound on the minimum required redundancy for fault-tolerant quantum circuits with classical inputs and outputs. Recent results show that due to physical resource constraints in quantum circuits, increasing redundancy can increase noise, which in turn may render many fault-tolerance schemes useless. So it is of both practical and theoretical interest to characterize the effect of resource constraints on the fundamental limits of fault-tolerant quantum circuits. Thus as an application of our lower bound, we characterize the fundamental limit of fault-tolerant quantum circuits with classical inputs and outputs under resource constraint-induced noise models.

quant-ph↗

arXiv2023

Efficient Equivariant Transfer Learning from Pretrained Models

Efficient transfer learning algorithms are key to the success of foundation models on diverse downstream tasks even with limited data. Recent works of Basu et al. (2023) and Kaba et al. (2022) propose group averaging (equitune) and optimization-based methods, respectively, over features from group-transformed inputs to obtain equivariant outputs from non-equivariant neural networks. While Kaba et al. (2022) are only concerned with training from scratch, we find that equitune performs poorly on equivariant zero-shot tasks despite good finetuning results. We hypothesize that this is because pretrained models provide better quality features for certain transformations than others and simply averaging them is deleterious. Hence, we propose λ-equitune that averages the features using importance weights, λs. These weights are learned directly from the data using a small neural network, leading to excellent zero-shot and finetuned results that outperform equitune. Further, we prove that λ-equitune is equivariant and a universal approximator of equivariant functions. Additionally, we show that the method of Kaba et al. (2022) used with appropriate loss functions, which we call equizero, also gives excellent zero-shot and finetuned performance. Both equitune and equizero are special cases of λ-equitune. To show the simplicity and generality of our method, we validate on a wide range of diverse applications and models such as 1) image classification using CLIP, 2) deep Q-learning, 3) fairness in natural language generation (NLG), 4) compositional generalization in languages, and 5) image classification using pretrained CNNs such as Resnet and Alexnet.

cs.LG↗

arXiv2023

Transformers are Universal Predictors

We find limits to the Transformer architecture for language modeling and show it has a universal prediction property in an information-theoretic sense. We further analyze performance in non-asymptotic data regimes to understand the role of various components of the Transformer architecture, especially in the context of data-efficient training. We validate our theoretical analysis with experiments on both synthetic and real datasets.

cs.LG↗

arXiv2023

Network Analysis as a Tool for Shaping Conservation and Development Policy: A Case Study of Timber Market Optimization in India

The incorporation of trees on farms can help to improve livelihoods and build resilience among small-holder farmers in developing countries. On-farm trees can help gen- erate additional income from commercial tree harvest as well as contribute significant environmental benefits and ecosystem services to increase resiliency. Long-term benefits from tree-based livelihoods, however, depend on sustainable timber harvesting. In this paper, we discuss the potential for network analysis as a tool to inform conservation and development decision-making. Specifically, we formulate the commercial tree market between farmers and traders as a transportation problem and optimize the transactions. We create a model of the commercial tree market in the Bilaspur district of Himachal Pradesh, India based on a detailed dataset of market interactions between farmers and timber traders, using the existing road network of this region. Using this model, we perform a maximum-flow-minimum-cost optimization for tree commodity flow. We compare the results of our optimized model with actual flow within the network, and we find a high potential to increase efficiency of market transactions within this region, noting a significant reduction to the minimum- cost flow value for our optimized model compared to the flow cost for actual transactions. We propose that using this network flow optimization model to strategically distribute permits can reduce costs associated with market transactions. Our results suggest that this direct policy action would be beneficial to the region. Finally, we suggest that cost savings could be used to establish tree planting programs to support a long-term sustainable tree market. Shaping policies to address these market inefficiencies in developing regions could help support and elevate tree-based livelihoods for farmers, traders, and industries.

cs.SI↗

arXiv2023

Learning Optimal Features via Partial Invariance

Learning models that are robust to distribution shifts is a key concern in the context of their real-life applicability. Invariant Risk Minimization (IRM) is a popular framework that aims to learn robust models from multiple environments. The success of IRM requires an important assumption: the underlying causal mechanisms/features remain invariant across environments. When not satisfied, we show that IRM can over-constrain the predictor and to remedy this, we propose a relaxation via $\textit{partial invariance}$. In this work, we theoretically highlight the sub-optimality of IRM and then demonstrate how learning from a partition of training domains can help improve invariant models. Several experiments, conducted both in linear settings as well as with deep neural networks on tasks over both language and image data, allow us to verify our conclusions.

cs.LG↗

arXiv2023

Equi-Tuning: Group Equivariant Fine-Tuning of Pretrained Models

We introduce equi-tuning, a novel fine-tuning method that transforms (potentially non-equivariant) pretrained models into group equivariant models while incurring minimum $L_2$ loss between the feature representations of the pretrained and the equivariant models. Large pretrained models can be equi-tuned for different groups to satisfy the needs of various downstream tasks. Equi-tuned models benefit from both group equivariance as an inductive bias and semantic priors from pretrained models. We provide applications of equi-tuning on three different tasks: image classification, compositional generalization in language, and fairness in natural language generation (NLG). We also provide a novel group-theoretic definition for fairness in NLG. The effectiveness of this definition is shown by testing it against a standard empirical method of fairness in NLG. We provide experimental results for equi-tuning using a variety of pretrained models: Alexnet, Resnet, VGG, and Densenet for image classification; RNNs, GRUs, and LSTMs for compositional generalization; and GPT2 for fairness in NLG. We test these models on benchmark datasets across all considered tasks to show the generality and effectiveness of the proposed method.

cs.LG↗

arXiv2022

Hypergraph-based Source Codes for Function Computation Under Maximal Distortion

This work investigates functional source coding problems with maximal distortion, motivated by approximate function computation in many modern applications. The maximal distortion treats imprecise reconstruction of a function value as good as perfect computation if it deviates less than a tolerance level, while treating reconstruction that differs by more than that level as a failure. Using a geometric understanding of the maximal distortion, we propose a hypergraph-based source coding scheme for function computation that is constructive in the sense that it gives an explicit procedure for finding optimal or good auxiliary random variables. Moreover, we find that the hypergraph-based coding scheme achieves the optimal rate-distortion function in the setting of coding for computing with side information and achieves the Berger-Tung sum-rate inner bound in the setting of distributed source coding for computing. It also achieves the El Gamal-Cover inner bound for multiple description coding for computing and is optimal for successive refinement and cascade multiple description problems for computing. Lastly, the benefit of complexity reduction of finding a forward test channel is shown for a class of Markov sources.

cs.IT↗

arXiv2022

Debiased Large Language Models Still Associate Muslims with Uniquely Violent Acts

Recent work demonstrates a bias in the GPT-3 model towards generating violent text completions when prompted about Muslims, compared with Christians and Hindus. Two pre-registered replication attempts, one exact and one approximate, found only the weakest bias in the more recent Instruct Series version of GPT-3, fine-tuned to eliminate biased and toxic outputs. Few violent completions were observed. Additional pre-registered experiments, however, showed that using common names associated with the religions in prompts yields a highly significant increase in violent completions, also revealing a stronger second-order bias against Muslims. Names of Muslim celebrities from non-violent domains resulted in relatively fewer violent completions, suggesting that access to individualized information can steer the model away from using stereotypes. Nonetheless, content analysis revealed religion-specific violent themes containing highly offensive ideas regardless of prompt format. Our results show the need for additional debiasing of large language models to address higher-order schemas and associations.

cs.CL↗