Distributing task-related neural activity across a cortical network through task-independent connections Nature Communications

For instance, we refer to a cell as circular that is characterized by an elliptical equation, but we don’t need to specify the radius of the circle. To take advantage of these patterns, we reparameterize the learned formula by making the fixed constants trainable. Such neurons will perform better than preset neurons since considering the complexity of tasks, there should be no one-size-fits-all neurons. By reparameterizing the learned formula, we can fine-tune the neuron’s behavior to better fit the task at hand.

Task area of neural networks

Our task is centered on global representations, so explicitly learning this attribute also tends to improve performance. Our node representations also seem to be more useful than edge representations, which makes sense since more information is loaded in these attributes. In this view all graph attributes have learned representations, so we can leverage them during pooling by conditioning the information of our attribute of interest with respect to the rest. For example, for one node we can consider information from neighboring nodes, connected edges and the global information.

Recent work on MTL for Deep Learning

In58 the authors had to break the EI balance in order to achieve non-linear computations. With our training procedure, individual neurons can be trained to perform complex tasks, such as generating the spiking activity of cortical neurons, without leaving the balanced regime. The work by59,60 trained all the recurrent weights of the dynamically balanced spiking networks. To maintain strong excitatory-inhibitory activities after training, they considered weight regularizations that constrained the trained weights close to strong initial EI weights. Instead, in our training setup, the strong initial EI connections were left unchanged throughout training, thus always provided the strong excitation and inhibition. In this study, we computationally explored the amount of subset training by varying (1) the number of trainable neurons and (2) the number of plastic synapses to the trained neurons.

Task area of neural networks

Next, we asked how the spreading of trained activity may depend on the type of neurons being trained. To address this question, we considered two training scenarios where either the excitatory or the inhibitory subnetwork (but not both) was trained to generate the target activity patterns (Fig. 4A, right). This analysis suggested that the ramping mode was the dominant component of the trained activity that was transferred to the untrained inhibitory neurons and shared with the fast-spiking ALM neurons.

No One-Size-Fits-All Neurons: Task-based Neurons for Artificial Neural Networks

What image features is an object recognizer looking at, and how does it piece them together into the distinctive visual signatures of cars, houses, and coffee cups? Looking at the weights of individual connections won’t answer that question. As mentioned before, MTL can be used to learn features that might not be easy to learn just using the original task. An effective way to achieve this is to use hints, i.e. predicting the features as an auxiliary task. Starting at the other extreme, [39] propose a bottom-up approach that starts with a thin network and dynamically widens it greedily during training using a criterion that promotes grouping of similar tasks. The widening procedure, which dynamically creates branches can be seen in Figure 4.

  • This is different from our work, in which we train only a subset of the neurons and investigate the role of untrained synapses in spreading the trained activity to untrained neurons.
  • With all the various inputs, we can start to plug in values into the formula to get the desired output.
  • Such neurons will perform better than preset neurons since considering the complexity of tasks, there should be no one-size-fits-all neurons.
  • The process in which the algorithm adjusts its weights is through gradient descent, allowing the model to determine the direction to take to reduce errors (or minimize the cost function).

In MTL for computer vision, approaches often share the convolutional layers, while learning task-specific fully-connected layers. This approach, however, still relies on a pre-defined structure for sharing, which may be adequate for well-studied computer vision problems, but prove error-prone for novel tasks. The constraints used for soft parameter sharing in deep neural networks have been greatly inspired by regularization techniques for MTL that have been developed for other models, which we will soon discuss. In this section, we compare the task-based networks with other state-of-the-art models over two real-world tasks. To highlight the superiority of the network using task-based neurons, we select advanced machine learning models for comparison, namely XGBoost[35], LightGBM[36], CatBoost[37], TabNet[38], TabTransformer[39], FT-Transformer[16] and DANETs[40]. All these models are either classic models or recent models that were published in prestigious venues of machine learning.

What is a neural network? A computer scientist explains

The trained activity failed to spread in networks of neurons that were not strongly coupled. Optogenetic perturbation experiments of ALM activity provided additional evidence that the ALM network is strongly coupled, supporting the applicability of the proposed mechanism for spreading the trained activity to cortical networks. Starting from this asynchronous state, the goal of training was to produce structured spiking rate patterns in a subset of neurons selected from the network. Specifically, our training scheme modified the recurrent and external plastic synapses projecting to the selected neurons, so that they generated target activity patterns when evoked by a brief external stimulus.

While useful in many scenarios, hard parameter sharing quickly breaks down if tasks are not closely related or require reasoning on different levels. Recent approaches have thus looked towards learning what to share and generally outperform hard parameter sharing. In addition, giving our models the capacity to learn a task hierarchy is helpful, particularly in cases that require different granularities. The model, which can be seen in Figure 8, allows to learn what layers and subspaces should be shared, as well as at what layers the network has learned the best representations of the input sequences.

A Joint Many-Task Model

It suggests that task-related activity observed in cortical regions during behavior can emerge from sparse synaptic reorganization to a subset of neurons and then propagate to the rest of the network through the strong, task-independent synapses. We applied our modeling framework to study the spread of task-related activity in the anterior lateral motor cortex (ALM) of mice performing a memory-guided decision-making task21. Similarly to neurons in the primate motor cortex28,29,30, the activity of many neurons in ALM ramps slowly during motor preparation and is selective to future actions21,31,32. These task-related activity patterns are widely distributed across the ALM and are highly heterogeneous across neurons. Large-scale measurements of neural activity show that learning can rapidly change the activity of many neurons, resulting in widespread changes in task-related neural activity1,2,3,4,5. For instance, a goal-directed behavior involving motor planning leads to widespread changes across the motor cortex1.

Unlike image and text data, social networks do not have identical adjacency matrices. Graphs are a useful tool to describe data you might already be familiar with. In these examples, the number of neighbors to each node is variable (as opposed to the fixed neighborhood size of images and text). We can digitize text by associating indices to each character, word, or token, and representing text as a sequence of these indices.

Similarly, when the inhibitory subpopulation was trained, the presynaptic inhibitory neurons were sampled from other trained inhibitory neurons while the presynaptic excitatory neurons were sampled from the entire excitatory population. The plastic weights from the external neurons to each trained neuron were trained by the learning algorithm. Other studies showed that a larger number of synapses across the entire network can be trained successfully, as long as they are weaker than the strong pre-existing random connections19,35,61. Specifically, several recent studies showed that it is possible to train networks to perform tasks by training weak presynaptic inputs, while constraining their connectivity to be of low-rank62,63.

Task area of neural networks

Each node is processed independently, as is each edge, as well as the global context. Now we’ve demonstrated that we can build a simple GNN model, and make binary predictions by routing information between different parts of the graph. This pooling technique will serve as a building block for constructing more sophisticated GNN models.

For example, we can consider a network of molecules, where a node represents a molecule and an edge is shared between two molecules if we have a way (reaction) of transforming one to the other . In this case, we can learn on a nested graph by having a GNN that learns representations at the molecule level and another at the reaction network how to use neural network level, and alternate between them during training. We can then build graphs by treating these objects as nodes, and their relationships as edges. Machine learning models, programming code and math equations can also be phrased as graphs, where the variables are nodes, and edges are operations that have these variables as input and output.

Task area of neural networks

Leave a Comment

Your email address will not be published. Required fields are marked *

Shopping Cart
Ncesitas ayuda?
Scroll to Top