Input Array Research Articles

Audio Deep Fake Detection with Sonic Sleuth Model

Information dissemination and preservation are crucial for societal progress, especially in the technological age. While technology fosters knowledge sharing, it also risks spreading misinformation. Audio deepfakes—convincingly fabricated audio created using artificial intelligence (AI)—exacerbate this issue. We present Sonic Sleuth, a novel AI model designed specifically for detecting audio deepfakes. Our approach utilizes advanced deep learning (DL) techniques, including a custom CNN model, to enhance detection accuracy in audio misinformation, with practical applications in journalism and social media. Through meticulous data preprocessing and rigorous experimentation, we achieved a remarkable 98.27% accuracy and a 0.016 equal error rate (EER) on a substantial dataset of real and synthetic audio. Additionally, Sonic Sleuth demonstrated 84.92% accuracy and a 0.085 EER on an external dataset. The novelty of this research lies in its integration of datasets that closely simulate real-world conditions, including noise and linguistic diversity, enabling the model to generalize across a wide array of audio inputs. These results underscore Sonic Sleuth’s potential as a powerful tool for combating misinformation and enhancing integrity in digital communications.

Coordinated Regulation of Renal Glucose Reabsorption and Gluconeogenesis by mTORC2 and Potassium.

The kidney proximal tubule is uniquely responsible for reabsorption of filtered glucose and gluconeogenesis (GNG). Insulin stimulates glucose transport and suppresses GNG in the proximal tubule, however, the signaling mechanisms and coordinated regulation of these processes remain poorly understood. The kinase complex mTORC2 is critical for regulation of growth, metabolism, solute transport, and electrolyte homeostasis in response to a wide array of inputs. Here we examined its role in the regulation of renal glucose reabsorption and GNG. Rictor, an essential component of mTORC2, was knocked out using the Pax8-LC1 system to generate inducible tubule specific Rictor knockout (TRKO) mice. These animals were subjected to fasting, refeeding, and variation in dietary K + . Metabolic parameters including glucose homeostasis and renal function were assessed in balance cages. Kidneys and livers were also harvested for molecular analysis of gluconeogenic enzymes, mTORC2-regulated targets, and plasma membrane glucose transporters. On a normal chow diet, TRKO mice had marked glycosuria despite indistinguishable blood glucose relative to WT controls. Kidney plasma membrane showed lower SGLT2 and SGLT1 in the fed state, supporting reduced renal glucose reabsorption. Additional metabolic testing provided evidence for renal insulin resistance with elevated fasting insulin, impaired pyruvate tolerance, elevated hemoglobin A1c, and increased renal gluconeogenic enzymes in the fasted and fed states. These effects were correlated with reduced downstream phosphorylation of Akt and the transcription factor FOXO4, identifying a novel role of FOXO4 in the kidney. Interestingly, high dietary K + prevented glycosuria and excessive GNG in TRKO mice, despite persistent reduction in mTORC2 substrate phosphorylation. Renal tubule mTORC2 is critical for coordinated regulation of sodium-glucose cotransport by SGLT2 and SGLT1 as well as renal GNG. Dietary K + promotes glucose reabsorption and suppresses GNG independently of insulin signaling and mTORC2, potentially providing an alternative signaling mechanism in states of insulin resistance. The kidney contributes to regulation of blood glucose through reabsorption of filtered glucose and gluconeogenesis. This study shows that mTORC2 and dietary potassium coordinate the regulation of sodium-glucose cotransport and glucose production in the kidney via independent mechanisms. New insights into the regulation of these processes in the kidney offer promising implications for diabetes mellitus management and treatment.

Real-time inference of compacted soil properties using deflection tests: An AI-driven solution informed by unsaturated soil mechanics principles

This paper presents a novel Artificial Intelligence (AI)-driven tool designed to convert deflection test results into crucial soil parameters essential for quality assurance in compaction projects. The accurate determination of these parameters, such as density and void ratio, is imperative for ensuring the structural integrity of infrastructures constructed on such soils. Moreover, it facilitates the utilization of modern non-destructive equipment in compaction endeavors. The determination of these parameters is notably challenging in unsaturated soils owing to the intricate interplay among factors such as suction, moisture content, void ratio, and resulting deflection.This paper presents a pioneering tool to address these challenges. By integrating unsaturated soil mechanics with advanced AI techniques, particularly reinforcement learning, the tool leverages a diverse array of inputs, including in-situ data, experimental observations, and physics-based modeling. This integration enables dynamic adaptation to changing field conditions and ensuring the tool’s real-time adaptability and predictive accuracy.Field trials validated the tool’s efficacy in predicting soil properties accurately without direct measurements of moisture content or suction, variables often unmeasured in practical soil compaction projects. This unique capability underscores significant advancements in real-time assessment of unsaturated soils, illustrating the transformative potential of AI in geotechnical engineering and unsaturated soil mechanics.

Bidirectional Optical Neural Networks Based on Free-Space Optics Using Lens Arrays and Spatial Light Modulator.

This paper introduces a novel architecture-bidirectional optical neural network (BONN)-for providing backward connections alongside forward connections in artificial neural networks (ANNs). BONN incorporates laser diodes and photodiodes and exploits the properties of Köhler illumination to establish optical channels for backward directions. Thus, it has bidirectional functionality that is crucial for algorithms such as the backpropagation algorithm. BONN has a scaling limit of 96 × 96 for input and output arrays, and a throughput of 8.5 × 1015 MAC/s. While BONN's throughput may rise with additional layers for continuous input, limitations emerge in the backpropagation algorithm, as its throughput does not scale with layer count. The successful BONN-based implementation of the backpropagation algorithm requires the development of a fast spatial light modulator to accommodate frequent data flow changes. A two-mirror-like BONN and its cascaded extension are alternatives for multilayer emulation, and they help save hardware space and increase the parallel throughput for inference. An investigation into the application of the clustering technique to BONN revealed its potential to help overcome scaling limits and to provide full interconnections for backward directions between doubled input and output ports. BONN's bidirectional nature holds promise for enhancing supervised learning in ANNs and increasing hardware compactness.

A new approach to Mergesort algorithm: Divide smart and conquer

It is well known that comparison-based algorithms cannot run faster than O(nlogn). Therefore the running times of Mergesort, Heapsort, and Quicksort algorithms are asymptotically optimal. As well as asymptotic running times, how fast sorting algorithms work in practice has gained importance lately. One algorithm is preferred, if it is slightly more efficient than another, due to the rapid increase in data generation. In this study, we consider the Mergesort algorithm with a different approach. Unlike the Mergesort algorithm, we do not perform unconditional division. Instead, we divide the input array into ascending and descending sub-arrays. Then, we merge the sub-arrays obtained as a result of the division by slightly modifying the classical Merge function. As a result of these two operations, we obtain three new sorting algorithms depending on how we get ascending and descending sub-arrays. Among these three algorithms, the third one, which is our main proposed algorithm, is more important both theoretically and practically. In this algorithm, we divide the input array into alternately ascending and descending sub-arrays. Moreover, the terms of the sub-arrays do not have to be consecutive terms of the input array. The asymptotic running time of our proposed algorithm is O(nlogn) in the worst case and O(n) in the best case. In practice, the proposed algorithm performs better than the classical Mergesort algorithm in arrays with Gaussian and Uniform distributions of various sizes. In an array of four million, it provided a 17% improvement in the Gaussian distribution compared to the classical Mergesort algorithm. In addition, it provided a 29% improvement in the Uniform distribution of the same size. The proposed algorithm performs also much better than traditional sorting algorithms in real data collected from various sources.

Abstract 2327: MADVAR: An algorithm that improves the relevance of computational biology output, while reducing compute time and space requirements

Abstract High-throughput methods implemented in biology research produce a continuously growing array of data input that are used to produce data output with an increasing abundance of features. While growth in the volume and diversity of data input can be highly valuable for studying biological systems, it presents the challenge of managing enormous quantities of features, many of which are not relevant to the specific research question being asked. This excess data input burdens storage and computation of downstream clustering and machine learning tasks. A common approach used to manage this data input relies on filters applied to the features by their variance across the sample set, while applying random cutoffs. Our proprietary algorithm (“MADVAR”) enables the prioritization of variable features from high-throughput continuous data, by automatically finding an optimal cutoff for the distribution of the data. Based on the right-skew nature of biological data distribution, MADVAR finds and excludes the "0 variance peak" using the median of the distributions and the median absolute deviation (MAD). MADVAR enables a faster analysis with a reduced memory requirement, and dramatically improves clustering results with minimal loss of relevant features. Citation Format: Gilad Silberberg, Michael Ritchie. MADVAR: An algorithm that improves the relevance of computational biology output, while reducing compute time and space requirements [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 2327.

Scalable Optical Convolutional Neural Networks Based on Free-Space Optics Using Lens Arrays and a Spatial Light Modulator.

A scalable optical convolutional neural network (SOCNN) based on free-space optics and Koehler illumination was proposed to address the limitations of the previous 4f correlator system. Unlike Abbe illumination, Koehler illumination provides more uniform illumination and reduces crosstalk. The SOCNN allows for scaling of the input array and the use of incoherent light sources. Hence, the problems associated with 4f correlator systems can be avoided. We analyzed the limitations in scaling the kernel size and parallel throughput and found that the SOCNN can offer a multilayer convolutional neural network with massive optical parallelism.

Reconstructing parameterized strings from parameterized suffix and LCP arrays

Reconstructing input from a data structure entails determining whether an instance of the data structure is in fact valid or not, and if valid, discovering the underlying data that it represents. In this paper we consider the parameterized suffix array (pSA) along with its corresponding parameterized longest-common-prefix (pLCP) array and solve the following problem. Given two arrays of numbers as input, A and P, does there exist a parameterized string S such that A is its pSA and P is its pLCP array? If the answer is positive, our algorithm produces a string S whose pSA is A and whose pLCP array is P. Although the naive approach would have to consider an exponential number of possibilities for such a string S, our algorithm's time complexity is only O(n2) for input arrays of size n.

High-spatial density snapshot imaging spectrometer enabled by 2-photon fabricated custom fiber bundles.

We report on a proof-of-concept snapshot imaging spectrometer developed using an array of optical fibers fabricated with 2-photon polymerization (2PP). The dense input array maps to an output array with engineered void spaces for spectral information. Previously, the development and fabrication of custom fiber arrays for imaging spectrometers have been a complex, time-consuming, and costly process, requiring a semi-manual assembly of commercial components. This work applies an automatic development process based on 2PP additive manufacturing with the Nanoscribe GmbH Quantum X system. The technique allows printing of arbitrary optical quality structures with submicron resolution with less than 5 nm roughness, enabling small core fibers/integrated arrays. Specifically, we developed an array prototype of 40 × 80 with 6-micron pitch at the input and 80-micron pitch at the output. The air-clad fibers had a core diameter of 5 µm. Fabricated optical fiber arrays were incorporated into a prism-based imaging spectrometer system with 48 spectral channels to demonstrate multi-spectral imaging. Imaging of a USAF target and color printed letter C as well as spectral comparisons to a commercial spectrometer were used to validate the performance of the system. These results clearly demonstrate the functionality and potential applications of the 3D-printed fiber-based snapshot imaging spectrometer.

Neural sensory stimulation does not interfere with the H-reflex in individuals with lower limb amputation.

Individuals with lower limb loss experience an increased risk of falls partly due to the lack of sensory feedback from their missing foot. It is possible to restore plantar sensation perceived as originating from the missing foot by directly interfacing with the peripheral nerves remaining in the residual limb, which in turn has shown promise in improving gait and balance. However, it is yet unclear how these electrically elicited plantar sensation are integrated into the body's natural sensorimotor control reflexes. Historically, the H-reflex has been used as a model for investigating sensorimotor control. Within the spinal cord, an array of inputs, including plantar cutaneous sensation, are integrated to produce inhibitory and excitatory effects on the H-reflex. In this study, we characterized the interplay between electrically elicited plantar sensations and this intrinsic reflex mechanism. Participants adopted postures mimicking specific phases of the gait cycle. During each posture, we electrically elicited plantar sensation, and subsequently the H-reflex was evoked both in the presence and absence of these sensations. Our findings indicated that electrically elicited plantar sensations did not significantly alter the H-reflex excitability across any of the adopted postures. This suggests that individuals with lower limb loss can directly benefit from electrically elicited plantar sensation during walking without disrupting the existing sensory signaling pathways that modulate reflex responses.

Enhancing Visual Feedback Control through Early Fusion Deep Learning.

A visual servoing system is a type of control system used in robotics that employs visual feedback to guide the movement of a robot or a camera to achieve a desired task. This problem is addressed using deep models that receive a visual representation of the current and desired scene, to compute the control input. The focus is on early fusion, which consists of using additional information integrated into the neural input array. In this context, we discuss how ready-to-use information can be directly obtained from the current and desired scenes, to facilitate the learning process. Inspired by some of the most effective traditional visual servoing techniques, we introduce early fusion based on image moments and provide an extensive analysis of approaches based on image moments, region-based segmentation, and feature points. These techniques are applied stand-alone or in combination, to allow obtaining maps with different levels of detail. The role of the extra maps is experimentally investigated for scenes with different layouts. The results show that early fusion facilitates a more accurate approximation of the linear and angular camera velocities, in order to control the movement of a 6-degree-of-freedom robot from a current configuration to a desired one. The best results were obtained for the extra maps providing details of low and medium levels.

Solar horizontal flow evaluation using neural network and numerical simulations with snapshot data

Abstract We suggest a method that evaluates the horizontal velocity in the solar photosphere with easily observable values using a combination of neural network and radiative magnetohydrodynamics simulations. All three-component velocities of thermal convection on the solar surface have important roles in generating waves in the upper atmosphere. However, the velocity perpendicular to the line of sight (LoS) is difficult to observe. To deal with this problem, the local correlation tracking (LCT) method, which employs the difference between two images, has been widely used, but this method has several disadvantages. We develop a method that evaluates the horizontal velocity from a snapshot of the intensity and the LoS velocity with a neural network. We use data from numerical simulations for training the neural network. While two consecutive intensity images are required for LCT, our network needs just one intensity image at only a specific moment for input. From these input arrays, our network outputs a same-size array of a two-component velocity field. With only the intensity data, the network achieves a high correlation coefficient between the simulated and evaluated velocities of 0.83. In addition, the network performance can be improved when we add LoS velocity for input, enabling us to achieve a correlation coefficient of 0.90. Our method is also applied to observed data.

The predictive management in campus heating system based on deep reinforcement learning and probabilistic heat demands forecasting

As a promising technology for replacing the rule-based decision-making in region heating systems (RHS), deep reinforcement learning (DRL) is a practical solution to identify the optimal control for heating equipment. However, as residential customers perform more casual energy-consumption behaviors, the intermittency and volatility of heat demands make managing heat supply and storage much harder for DRL agents. This study proposes a novel predictive management method for campus heating systems (CHS) with air-source heat pumps (AHP) and thermostatic water tanks. The novelty of the proposed method lies in the combination of the heat demands forecasting model and DRL-based adaptively controlling for heat supply equipment, which is firstly proposed to improve the heating supply reliability and reduce the storage dependence for CHS. Specifically, an enhanced rule, namely minimum length hamming encoding, and an input array constructing method is introduced to deal with discrete feature data and then improve the accuracy of deterministic heat demands forecasting based on long-short term memory (LSTM), and the Kernel density estimation (KDE) are employed to obtain the prediction intervals (PIs) from H-step ahead heat demands forecasting series. Followed by these, the twin delayed deep deterministic policy gradient, a model-free DRL control algorithm, is adopted for adaptively adjusting the output flow rate of AHP and then the storage of the hot water tank. To demonstrate the validity of the proposed method, a case study is presented where a campus heat demands forecasting achieves a maximum accuracy gain of 4.52%, and an optimal AHP operating controlling determined from PIs achieves a better cost reduction and supply reliability, which is superior over the conventional method using real-time heat demands or deterministic forecasting results as input.

Colonial origins and quality of education evidence from cameroon

We revisit the question of colonial legacies in education by focusing on quality rather than quantity. We study Cameroon, a country where a Francophone education system with French colonial origins coexists with an Anglophone system with British colonial origins. This allows us to investigate the impact of different teaching practices on students’ test scores. We find that pupils schooled in the Francophone system perform better in mathematics in Grade 5, with test scores higher by two thirds of a standard deviation. Thanks to detailed school survey data, we are able to account for a wide array of inputs of the education production function, such as the economic and social conditions of students, the material conditions of the schools and classrooms, as well as some information on the teachers’ practices and pedagogical culture. We find that Francophone schools have better classroom equipment and that Francophone teachers use more vertical teaching methods, but that these differences cannot explain why Francophone students perform better in mathematics. In the end, we cannot pin down the exact mechanism behind our result.

Crossmodal interactions in human learning and memory.

Most studies of memory and perceptual learning in humans have employed unisensory settings to simplify the study paradigm. However, in daily life we are often surrounded by complex and cluttered scenes made up of many objects and sources of sensory stimulation. Our experiences are, therefore, highly multisensory both when passively observing the world and when acting and navigating. We argue that human learning and memory systems are evolved to operate under these multisensory and dynamic conditions. The nervous system exploits the rich array of sensory inputs in this process, is sensitive to the relationship between the sensory inputs, and continuously updates sensory representations, and encodes memory traces based on the relationship between the senses. We review some recent findings that demonstrate a range of human learning and memory phenomena in which the interactions between visual and auditory modalities play an important role, and suggest possible neural mechanisms that can underlie some surprising recent findings. We outline open questions as well as directions of future research to unravel human perceptual learning and memory.

IMPLEMENTATION OF THE INTELLECTUAL SYSTEM OF SENTIMENT ANALYSIS AND CLUSTERIZATION OF PUBLICATIONS IN THE TWITTER SOCIAL NETWORK

Thanks to the intensive development of social networks, the intensity of exchange of short electronic text messages is constantly increasing, the tone of which can serve as a sensitive indicator of public mood and important social phenomena, interesting for sociologists, politicians, economists, and specialists in other fields. In this regard, the task of automating the processing of such natural language messages is of significant scientific and practical interest. The object of this study is the sentiment of user publications in the Twitter social network. Due to the great popularity of the social network itself and the large number of user messages, which are short in nature, it is possible to conveniently determine the mood of user posts and combine them into clusters according to the given parameters of the intelligent system. The subject of the study is methods and algorithms for analysing the sentiment of large arrays of messages containing the necessary keywords and relating to a certain specific topic, determining the factors and distributions of the sentiment of messages based on the input array of system data, dividing messages into main groups and providing estimates within certain defined limits in to each group, division into clusters according to the obtained search point and display of the obtained results in the desired format. The purpose of the work is to implement an intelligent system of sentiment analysis and clustering of publications based on a recurrent neural network of long short-term memory (LSTM) and the k-means clustering algorithm. The following main tasks are solved in the work: 1. To analyse the most used and newest algorithms, methods, approaches and means of implementing tasks of sentiment analysis and clustering of publications in social networks. 2. To develop a conceptual structure of an intellectual system of sentiment analysis and clustering of publications. 3. To form functional tasks for the key modules of the created intelligent system of sentiment analysis and clustering of publications in the Twitter social network. 4. Implement an intelligent system of sentiment analysis and clustering of publications based on a recurrent neural network and the k-means clustering algorithm and conduct experimental verification. Among the methods used for this purpose are the recurrent neural network of long short-term memory; k-means clustering algorithm. The following results were obtained: the general structure of the intellectual system of sentiment analysis and clustering of publications was analyzed, designed and built. The main task of creating the system, first of all, was to improve the recurrent neural network of long-short-term memory, which, thanks to the improved algorithm, significantly facilitates text processing by natural language processors according to text data of a certain size. Also, a special clustering algorithm, namely k-means, was used in parallel, thanks to which it was possible to change the general approach to clustering and the creation of final clusters, in accordance with the obtained results of the work of the recurrent neural network. Conclusions: As a result of applying a combination of LSTM neural network and k-means clustering algorithm, it was possible to speed up the process of sentiment analysis and clustering of posts in the Twitter social network by 10...15% compared to similar convolutional neural networks and hierarchical clustering.

Non-negative low-rank approximations for multi-dimensional arrays on statistical manifold

Although low-rank approximation of multi-dimensional arrays has been widely discussed in linear algebra, its statistical properties remain unclear. In this paper, we use information geometry to uncover a statistical picture of non-negative low-rank approximations. First, we treat each input array as a probability distribution using a log-linear model on a poset, where a structure of an input array is realized as a partial order. We then describe the low-rank condition of arrays as constraints on parameters of the model and formulate the low-rank approximation as a projection onto a subspace that satisfies such constraints, where parameters correspond to coordinate systems of a statistical manifold. Second, based on information-geometric analysis of low-rank approximation, we point out the unexpected relationship between the rank-1 non-negative low-rank approximation and mean-field approximation, a well-established method in physics that uses a one-body problem to approximate a many-body problem. Third, our theoretical discussion leads to a novel optimization method of non-negative low-rank approximation, called Legendre Tucker rank reduction. Because the proposed method does not use the gradient method, it does not require tuning parameters such as initial position, learning rate, and stopping criteria. In addition, the flexibility of the log-linear model enables us to treat the problem of non-negative multiple matrix factorization (NMMF), a variant of low-rank approximation with shared factors. We find the best rank-1 NMMF formula as a closed form and develop a rapid rank-1 NMF method for arrays with missing entries based on the closed form, called A1GM.

A scalable optical computer based on free-space optics using lens arrays and a spatial light modulator

A scalable optical computer based on free-space optics and Koehler illumination was proposed to obtain optical parallelism in an artificial neural network. The Koehler illumination scheme can provide more uniform illumination and less crosstalk compared with the previous systems based on Abbe illumination. The proposed design was realised in hardware with 2 × 2 inputs and 2 × 2 outputs using light-emitting diodes, a liquid crystal display, lens arrays, and a detector array assembled in an optical cage system with subsidiary electronics. The feasibility of the constructed optical computer was demonstrated by achieving the expected results of the NAND and NOR logics in parallel in the nondifference mode. Additionally, a simple pattern-checking function in the difference mode was shown. In addition, the limit of the scalability and the effect of geometric aberration was discussed. Furthermore, a clustering technique was suggested to overcome the limitations of scalability. Clustering can continue to increase the array size and throughput by creating a new block of doubled input and output arrays from multistacks of basic optical modules. The proposed architecture and clustering method have the potential to realise massive optical parallelism if large-array components and high-speed electronics are used in the future.

Improved Threat Detection in Walk-Through Security Scanning Using an Optimized Polarimetric Target Decomposition Method

Reliable and convenient walk-through security scanning, which doesn't separate people or impede their movement, is an extremely challenging task. In this paper, a novel approach for a security check with an overhead observation and a polarimetric target decomposition is presented. The viewing angle of the scanner equals a side-looking airborne radar. However, it will be shown that the established polarimetric target decomposition methods of remote-sensing are not well suited for close-range radar imaging and need to be adapted due to the differences in the geometry of the imaging scenario. The usage of a multiple input and multiple output (MIMO) array and the shorter distance between array and target need a changed decomposition technique in order to distinguish between persons with or without worn threat objects. The differences between radar remote-sensing and close-range imaging scenarios are investigated. An optimized version of the Sato four-component polarimetric scattering decomposition is derived. The proposed close-range adjustment is applied to the model and investigated experimentally with a 4-to-12 GHz fully polarimetric MIMO imaging system. Polarimetric decomposition is carried out on defined test structures with known scattering mechanisms, as well as on mannequins and persons with different threat objects like guns or explosives. In test campaigns, promising results were achieved for a correct target decomposition in radar-based walk-through security scanning.

Wdr59 promotes or inhibits TORC1 activity depending on cellular context

Target of Rapamycin Complex I (TORC1) is a central regulator of metabolism in eukaryotes that responds to a wide array of negative and positive inputs. The GTPase-activating protein toward Rags (GATOR) signaling pathway acts upstream of TORC1 and is comprised of two subcomplexes. The trimeric GATOR1 complex inhibits TORC1 activity in response to amino acid limitation by serving as a GTPase-activating protein (GAP) for the TORC1 activator RagA/B, a component of the lysosomally located Rag GTPase. The multi-protein GATOR2 complex inhibits the activity of GATOR1 and thus promotes TORC1 activation. Here we report that Wdr59, originally assigned to the GATOR2 complex based on studies performed in tissue culture cells, unexpectedly has a dual function in TORC1 regulation in Drosophila. We find that in the ovary and the eye imaginal disc brain complex, Wdr59 inhibits TORC1 activity by opposing the GATOR2-dependent inhibition of GATOR1. Conversely, in the Drosophila fat body, Wdr59 promotes the accumulation of the GATOR2 component Mio and is required for TORC1 activation. Similarly, in mammalian HeLa cells, Wdr59 prevents the proteolytic destruction of GATOR2 proteins Mio and Wdr24. Consistent with the reduced levels of the TORC1-activating GATOR2 complex, Wdr59KOs HeLa cells have reduced TORC1 activity which is restored along with GATOR2 protein levels upon proteasome inhibition. Taken together, our data support the model that the Wdr59 component of the GATOR2 complex functions to promote or inhibit TORC1 activity depending on cellular context.