Concepts
5,907Terms and methods explained by more than one lecturer. Each opens at the minute it is taught.
Engineering
127- BasisA set of linearly independent vectors that span a vector space.
- LinearizationApproximating a non-linear system with a linear system near a fixed point to determine its stability.
- Newton's methodAn iterative numerical method for finding successively better approximations to the roots of a real-valued function using tangent lines.
- Jacobian matrixA matrix of all first-order partial derivatives of a vector-valued function.
- Kalman filterAn optimal full-state estimator used to estimate the full state of a system from measurements and inputs.
- Eigenvalues and eigenvectorsA scalar lambda and a vector X such that multiplying the matrix Q by X yields lambda times X.
- ProjectionThe point along a line (or subspace) closest to a given vector.
- DimensionThe dimension of a space is the number of vectors in any basis for that space.
- Null spaceThe null space of a matrix is the set of all solutions to the homogeneous equation Ax = 0.
- Gaussian eliminationAn algorithm for solving systems of linear equations through a sequence of row operations.
- SparsityA structural property of a matrix where a large number of its entries are zero.
- Transition matrixA matrix used to describe the probabilities of transitions between states in a Markov chain.
- Branch and boundAn algorithm design paradigm for discrete and combinatorial optimization problems that systematically enumerates candidate solutions.
- Matrix exponentialThe matrix exponential is defined by the power series I + At + 1/2!A²t² + …, analogous to the scalar exponential series.
- Processing-in-memoryComputing inside the memory chips to process data close to where it is stored.
- ContrapositiveAn equivalent logical statement formed by negating both parts of an implication and reversing their order, specifically if Q is not true then P is not true.
- Linear quadratic regulatorAn optimal control problem for a linear system with a quadratic cost function.
- Temporal localityThe principle that if you access an address, you are very likely in the near future to access it again.
- Truth tableA table that shows the logical output of a circuit for each possible input combination.
- Impulse responseThe output of a dynamical system when presented with a brief input signal, characterizing the system's behavior.
- Model predictive controlA control method using a dynamic model of the system to predict and optimize future behavior over a rolling time horizon.
- PrefetchingA technique to automatically anticipate and load needed memory addresses before they are requested to reduce latency.
- Spatial localityAccessing memory locations close to recently accessed locations.
- BlockingFetching an entire block of data instead of a single word to amortize the latency cost over multiple items.
- Stochastic matrixA non-negative matrix where all columns sum to one, representing a linear transformation of a probability vector.
- ViscosityA fluid property describing internal friction and forces parallel to fluid flow against boundaries.
- Lyapunov functionA scalar function used to prove the stability of an equilibrium point of a dynamical system.
- Memory hierarchyA structure of different memory types in a computer organized by capacity, speed, and cost, resembling a funnel from large/slow storage to fast/small processor storage.
- Complementary slacknessA condition relating the optimal primal and dual solutions in linear programming where either a variable is zero or its corresponding constraint is tight.
- DecoderA combinational logic block that acts as an input pattern detector with n inputs and outputs where exactly one output is active for each input pattern.
- Mass matrixA positive definite matrix mapping generalized accelerations to generalized forces, representing the inertial properties of the system.
- Open-loop controlA control strategy where the control input is computed based on a model without using sensor measurements of the output to correct for errors.
- StrideThe distance between consecutive elements of a data structure accessed in memory.
- Trajectory optimizationA specific problem of selecting a sequence of states and actions that optimize some outcome.
- EllipsoidA four-dimensional region in defined by the inequality (π |z₁|²)/a + (π |z₂|²)/b ≤ 1.
- Invariant setA set of states such that any trajectory starting in the set remains in the set for all time.
- NullspaceThe nullspace of a matrix is the set of vectors that are mapped to zero.
- ReachabilityThe property that a state xi in can be reached from the origin by some prescribed control input u(t).
- Riccati equationA matrix equation whose solution gives the cost matrix S needed to determine the optimal control law in an LQR problem.
- Calcite compensation depthThe depth in the ocean below which the rate of supply of calcium carbonate is outpaced by the rate of dissolution, such that no calcite is preserved.
- Convex hullThe set of all convex combinations of points in a given set.
- DualityA foundational concept in optimization relating primal problems and their dual formulations through weights and constraints.
- Intersection of convex setsThe intersection of any two or more convex sets is always a convex set.
- Level setA set of points where a function takes on a constant value, corresponding to indifference curves in economics.
- OracleA black box that can answer membership queries for a specific language in a single step.
- Orthogonal transformationA linear transformation T that preserves lengths, meaning the norm of T(x) equals the norm of x for every vector x.
- Right inverseA matrix such that A A_right⁻¹ = I when the matrix has full row rank.
- Similarity transformationA transformation of a matrix that preserves its eigenvalues.
- State transition matrixA matrix that describes how the state vector of a system evolves from one time step to the next.
- Content addressable memoryA specialized memory type that searches its contents based on input data rather than a memory address, used in caches to match tags.
- Block eliminationA method for solving partitioned linear equations by eliminating variables corresponding to structured sub-blocks.
- Cholesky factorizationA factorization of a Hermitian positive-definite matrix into the product of a lower triangular matrix and its conjugate transpose.
- Convex optimization problemAn optimization problem where the objective and constraint functions are convex.
- Critical pathThe longest delay path in the combinational logic that determines the minimum clock cycle time.
- Forward kinematicsThe mapping from joint space to operational space, determining the position and orientation of the robot's end effector from given joint parameters.
- LU factorizationThe factorization of any nonsingular matrix into a permutation matrix, a lower triangular matrix, and an upper triangular matrix.
- ResidualThe vector , which equals the negative gradient of at .
- Single program multiple dataSPMD is what happens when you run the same program on each processing element, where each takes care of a different chunk of the problem split up by using the rank.
- Tikhonov regularizationA scalarized least-squares problem that adds a scaled norm penalty on the weights to regularize the solution.
- Bang-bang controlA control strategy that switches abruptly between maximum and minimum limits at all times.
- Binary encodingAn encoding scheme that uses the minimum possible number of bits to represent states.
- Cache missAn instance where requested data is not found in the CPU cache, forcing a slower fetch from main memory.
- CirculationAn element in the null space of the node incidence matrix representing a circulating flow in a network.
- Field programmable gate arrayAn FPGA is a software-reconfigurable hardware substrate containing reconfigurable functions, interconnections, and input/output interfaces.
- Hit rateThe percentage of memory accesses that successfully find their data in the cache rather than missing.
- Interior point methodAlgorithms that travel inside the feasible set rather than around the edges to solve optimization problems.
- Inverse kinematicsInverse kinematics gives you the required joint angles given a desired position for the hand.
- Jordan blockAn upper bidiagonal square matrix with repeated eigenvalues on the diagonal and ones on the superdiagonal.
- Linear quadratic GaussianA control problem with linear dynamics, quadratic cost functions, and additive Gaussian noise where the separation principle applies.
- Lookup tableA memory array used to perform logic functions where each address acts as a row in the truth table and each data bit stores the corresponding output value.
- Matrix normThe maximum gain of a matrix, defined as the square root of the largest eigenvalue of A transpose A.
- Spectral normThe maximum singular value of a matrix, defined as the square root of the largest eigenvalue of X transpose X.
- Angle between vectorsThe angle between two nonzero vectors is the inverse cosine of their dot product divided by the product of their lengths.
- BackpressurePausing operator execution when buffers are full or when waiting for remote I/O to prevent memory from ballooning.
- Backtracking line searchA line search method starting at t equals 1 and repeatedly multiplying by beta until the Armijo condition is satisfied.
- Basin of attractionThe set of initial conditions that lead to a particular attractor, such as a rolling or standing fixed point.
- Block matrix multiplicationA method of multiplying partitioned matrices in smaller sub-blocks that improves cache efficiency and performance.
- Chance constraintConstraints that bound the probability of constraint violation below some scalar value alpha.
- CliqueA set of vertices in a graph that are all pairwise connected with each other.
- Common sub-expression eliminationAn optimization that identifies identical sub-expressions in a query plan and computes them only once, sharing the result.
- Defective matrixA matrix that is not diagonalizable because it lacks a full set of linearly independent eigenvectors.
- DilatationStretching or scrunching something along the axes.
- EqualizationSubtracting a corrupted measurement component to compensate for distortion in a signal.
- Error budgetAn agreed-upon limit on how long a service can be unreliably functioning or down within a period.
- EustasyThe globally averaged value of sea level change, serving as a direct measure of ice volume.
- Execute phaseThe phase where a client transaction performs all its required reads and writes on primary servers.
- False negativeThe probability of selecting hypothesis 1 when the sample was actually generated by distribution 2.
- FormalizationFormalization involves the addition of more structure to what you are working on by clearly articulating why and how you are performing it.
- Freeze valveA valve that melts when the temperature increases to drain the fuel to a drain tank.
- FunnelA time-varying Lyapunov function region around a trajectory that guarantees convergence to a target set.
- Gram-Schmidt procedureAn algorithm that takes a set of independent vectors and generates an orthonormal basis spanning the same subspace.
- HapticsHaptics comes from the Greek word describing the sense of touch, allowing haptic interaction with virtual or real environments.
- HyperplaneA set of the form {x | } where a is non-zero.
- Hypothesis testingA statistical method to decide between multiple possible models or distributions based on observed samples.
- IrisAn iterative procedure that solves optimization problems to find big polytopes and then ellipses filling them up to generate big collision-free regions in configuration space.
- Jordan formThe nicest, most diagonal-like matrix in a family of similar matrices that shares the same eigenvalues.
- KKT matrixA block matrix that arises from the first-order optimality conditions of a constrained quadratic optimization problem.
- Large behavior modelsLarge-scale machine learning models trained on data to output robot behaviors and actions.
- Line searchA step in optimization that determines the positive step size t along a given search direction.
- Min-termsA canonical product term in Boolean logic that evaluates to 1 for a single combination of input variables.
- Minimal elementAn element in is minimal if there is no other element in distinct from that is smaller than .
- Molten salt reactorA nuclear reactor in which the primary nuclear fuel is dissolved in molten fluoride salt rather than water.
- Monic polynomialA single-variable polynomial in which the leading coefficient is equal to one.
- Observation modelA function that maps the state vector to a vector of observations.
- Open circuit voltageThe voltage measured across a circuit's terminals when no current is flowing through them.
- Optimal solutionAn optimal solution is a point satisfying all constraints that achieves the smallest objective value among all feasible vectors.
- Perspective functionA mathematical technique showing that non-negative scaling of a convex set or function preserves convexity.
- Polynomial time algorithmAn algorithm where the number of operations is bounded by a polynomial of the input size measured in bits.
- PolytopeA geometric object with flat sides formed by the intersection of half-spaces defined by linear inequalities.
- Rate controlA static optimization problem of assigning rates to flows in a network subject to link capacity constraints.
- Reachable setThe set of all states that can be reached from a given state or tree within a specific time bound or step.
- Receiver operating characteristicAn analysis method used to evaluate diagnostic methods where an AUC of 0.7 to 0.8 indicates a test with good diagnostic potential.
- Recursive least squaresAn algorithm to compute least-squares estimates sequentially as new measurements arrive over time without recomputing the entire sum.
- Ridge regressionAn optimization problem involving the minimization of the L2 norm subject to linear equality constraints.
- Robustness (margin)A property where two points with different labels must be separated by a minimum distance called the margin.
- RoundingQuickly searching and turning fractional convex relaxation solutions into discrete binary path solutions.
- S procedureA technique to test if an implication holds by adding a multiplier polynomial to fight negative values outside a region.
- Second-order cone programmingAn optimization problem that minimizes a linear objective function subject to second-order cone constraints.
- Semidefinite programA generalization of linear programs where the variables are symmetric matrices constrained to be positive semidefinite.
- Sherman-Morrison-Woodbury formulaA formula that computes the inverse of a rank-k perturbation of a matrix using the inverse of a smaller k-by-k matrix.
- Sparse LU factorizationThe ability to explicitly decompose a sparse matrix into the product of a lower triangular and an upper triangular matrix.
- State estimatorA system that estimates the internal states of a dynamical system from measurements of its outputs.
- Static schedulingThe process where the compiler schedules instructions at compile time rather than the hardware doing it at runtime.
- Stefan-Boltzmann lawThe mathematical relationship relating incoming stellar radiation to outgoing thermal radiation emitted by a planet.
- Support functionA function that maps a direction to the maximum value of the inner product of that direction with points in a convex set.
- Uncanny ValleyA psychological phenomenon where computer graphics or images that appear almost human, but not quite, elicit feelings of eeriness or revulsion.
- ZonotopeAn H-polyhedron that is the affine projection of the unit box.